DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Status of Claims
This is a final rejection in response to amendments/remarks filed on 12/03/2025. Claims 1, 7, 13, 19, and 20 are amended. Claims 10 and 11 are cancelled without prejudice. Therefore, claims 1-9 and 12-20 remain pending and are examined herein.
Priority
Acknowledgment is made of applicant’s priority claim to provisional Application No. 63/571,374, filed on 03/28/2024.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-9, and 12-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1: Is the claim to a Process, Machine, Manufacture, or Composition of Matter?
Claims 1-9, &12: A system, comprising: at least one processor; and at least one memory including program code which when executed by the at least one processor causes operations comprising:
Claims 13-19: A method comprising:
Claims 20: A non-transitory computer readable store medium including executable code which when executed by at least one processor causes operations comprising:
Claims 1-12 are directed to an apparatus with processors and memory devices which is an apparatus claims and falls under at least “machine or manufacture.” Claims 13-19 are directed to a method which falls under “process.” Claim 20 is directed to a non-transitory storage medium which falls under at least “manufacture.” Therefore, all of the claims fall under at least one potentially eligible subject matter category and are to be further analyzed under step 2.
Step 2a Prong 1: Is the claim reciting an Judicial Exception(A Law of Nature, a Natural Phenomenon (Product of Nature), or An Abstract Idea?)
The claims under the broadest reasonable interpretation in light of the specification are analyzed herein. Representative claims 1, 13 and 20 are marked up, isolating the abstract idea from additional elements, wherein the abstract idea is in bold and the additional elements have been italicized as follows:
Claim 1 Preamble: A system, comprising: at least one processor; and at least one memory including program code which when executed by the at least one processor causes operations comprising:
Claim 13 Preamble: A method comprising:
Claim 20 Preamble: A non-transitory computer readable store medium including executable code which when executed by at least one processor causes operations comprising:
Claim 1 Body (also representative of claims 9 and 10 body): receiving, from a user interface, a request including at least one material for which an emission factor suggestion is requested;
- providing the at least one material to a language model;
- in response to the providing, receiving, from the language model, at least one word embedding representative of the at least one material;
- initiating, via an application programming interface (API), a semantic search of a vector database storing a plurality of word embeddings representative of a plurality of materials each mapped to a corresponding emission factor stored in an emission factors database, wherein the semantic search comprises comparing, in a vector space defined by the plurality of word embeddings,
the at least one word embedding representative of the at least one material to at least a portion of the plurality of word embeddings;
- identifying, based on a similarity metric, at least one matching word embedding for the at least one word embedding representative of the at least one material, wherein the at least one matching word embedding maps to at least one emission factor stored in the emissions factor database;
- sending, to the user interface, a response including the at least one emission factor and a corresponding confidence score based on the similarity metric to indicate a similarity between the at least one word embedding and the at least one matching word embedding; and
- storing the at least one word embedding representative of the at least one material in the vector database with a key that maps the at least one word embedding to the at least one emission factor stored in the emission factors database.
When evaluating the bolded limitations of the claims under the broadest reasonable interpretation in light of the specification, it is clear that representative claims 1, 13 and 20 recite an abstract idea within the category of “certain methods of organizing human activity.” More specifically, the present invention falls under the sub-grouping “commercial or legal interactions” including agreements in the form of contracts, legal obligations, advertising, marketing or sales activities or behaviors, and business relations as outlined in MPEP 2106.04(a)(2)(II)(B). In the instant case, the claims in bold recite at least the steps of “using word embedding mappings to... identify, based on a similarity metric, at least one matching word embedding...which maps to at least one emission factor, sending a response including the at least one emission factor, and storing the embedding with a key mapped to an emission factor” Therefore, the abstract idea at hand is the determining of an emission factor of materials based on the similarity of the material to word embeddings of products with known emission factors. The background of the specification supports the examiner’s assertion that the claims fall under the commercial interaction of business/sales activity,
“[0002]To determine a carbon footprint (or other types of impact, such as land or water use, and/or the like) for an item, such a product that is being used or acquired by an entity such as a user or a company, the emission factor for each of the materials of the item is used. The phrase “emission factor” refers to a value, such as a coefficient, that can be used to determine greenhouse gas emissions. The emission factor for a given material may take into account for example production of a material, transport of the material, disposal of the material, and/or the like. Often, one or more third party databases must be searched to identify an emission factor for a given material, and if an emission factor is identified, the emission factor is manually mapped to the material.”
Estimating a coefficient such as an emission factor, which is a result of production, transport, and disposal of a material is merely a commercial interaction because it is used to inform a business decision, such as determining which material to use based on comparing the carbon footprints.
Even when considering the series of steps that have been added in the amendments, particularly initiating a semantic search of a vector database storing a plurality of word embeddings representative of a plurality of materials each mapped to a corresponding emission factor stored in an emission factors database, wherein the semantic search comprises, comparing, in a vector space defined by the plurality of word embeddings, the at least one word embedding to at least a portion of the plurality of word embeddings, the claims are still reciting “certain methods of organizing human activity.” In the manner in which the steps are claimed, the steps above are no more than mere instructions to an individual to perform data analysis steps, in which the analysis is recited at a level of generality such that it encompasses any manner of performing the semantic search on the existing database of word embeddings. Within the broadest reasonable interpretation, the scope of the limitation above allows for any way of performing basic semantic search of a vector database, word embedding, and similarity comparisons can be performed manually with a pen and paper. For example, an individual can perform a semantic search comprising comparing a word embedding to the other word embeddings in a vector space, such as merely comparing numbers to each other. The steps are not technically enough to be limited to technological implementation, because “comparing” is broad enough to cover any method of achieving the outcome, without defining a specific series of steps or mechanisms to arrive at the outcome. Therefore, the claims merely claim the idea of retrieving an emission factor via any use of semantic search on a vector database in comparing embeddings. At this level of generality the claims still fall under at least one abstract idea under “commercial or legal interactions,” because it merely recites interactions between an individual and a computer that fall within business activities. Since “emissions factor database” merely indicates the type or source of data, and the functions of “initiating a semantic search,” “comparing,” and “storing” are merely instructions to perform generic data interactions to achieve a ”certain method of organizing human activity,” the claims are still reciting an abstract idea.
Therefore, in view of the specification, and the claim language, the claims recite “certain methods of organizing human activity” under the sub-category “commercial or legal interactions” and are to be further analyzed under Prong 2.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
Claims 1, 13, and 20 recite the following the additional elements:
- A system, comprising: at least one processor; and at least one memory including program code which when executed by the at least one processor causes operations in claim 1
- A non-transitory computer readable store medium including executable code which when executed by at least one processor causes operations in claim 13
- user interface in claims 1, 13, 20
-language model in claims 1, 13, 20
The additional elements listed above, when considered individually and in combination with the claim as a whole, no more than a recitation of the words “apply it” (or an equivalent) or mere instructions to implement an abstract idea or other exception on generic computing components as outlined in MPEP 2106.05(f). In this case, the abstract idea of “using word embedding mappings to... identify, based on a similarity metric, at least one matching word embedding...which maps to at least one emission factor, sending a response including the at least one emission factor, and storing the embedding with a key mapped to an emission factor” is merely instructed to be performed on generic computing components such as a memory, processor, computer, a non-transitory storage medium, and a user interface. It is evident in at least paragraph [0066-0067] of the specification that these computer components are intended to be any computer capable of performing the functions, therefore, no improvement to a computer or a technical field has been implemented in the claims, which is one of the consideration in MPEP 2106.05(a). For example, the user interface is merely part of the client device, as an input/output device to receive and display data, which is merely operating in its ordinary capacity. No specific structural components such as the functionality of the user interface layout (buttons, sliders, or new input/output techniques) have been described, therefore the user interface is merely part of the generic computer.
Furthermore, the additional element “language model” is no more than a general link to the field of language models. The claims merely recite providing data to the language model and receiving data as an output from the language model but do not meaningfully limit the use of language models on the claims. See MPEP 2106.05(h) for more information on general links to technological environments, or fields of use. Even when considering the abstract idea elements used along with the language model, such as the semantic search, comparing embeddings in the vector space, and storing the embeddings with a key, these still fall within “apply it,” because they merely claim the idea of using a computer to perform the abstract idea wherein the series of steps are recited at such a high level of generality that they encompass any use of a general purpose computer to achieve the outcome. Therefore, in line with MPEP 2106.05(f), they are merely claiming an idea of an outcome, and in line with MPEP 2106.05(a) they fail to improve any computer functionalities or any technology.
Therefore, the additional elements, whether analyzed individually or as an ordered combination, fail to integrate the abstract idea into a practical application, even when considering the claims as a whole. Therefore, claims 1, 13, 20 are directed to an abstract idea.
Step 2B: Does the claim recite additional elements that amount to significantly more than the judicial exception?
Claims 1, 13, and 20 recite the following the additional elements:
- A system, comprising: at least one processor; and at least one memory including program code which when executed by the at least one processor causes operations in claim 1
- A non-transitory computer readable store medium including executable code which when executed by at least one processor causes operations in claim 13
- user interface in claims 1, 13, 20
-language model in claims 1, 13, 20
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception because, when considered separately and as an ordered combination, they do not add significantly more (also known as an “inventive concept”) to the exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using generic computing devices such as a memory, processor, computer, user interface and a non-transitory storage medium to perform the abstract idea of “using word embedding mappings to... identify, based on a similarity metric, at least one matching word embedding...which maps to at least one emission factor, sending a response including the at least one emission factor, and storing the embedding with a key mapped to an emission factor” amounts to no more than mere instructions to apply the exception using generic computing components. (See MPEP 2106.05(f)). Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. Furthermore, the use of “language models” is merely a general link to the technological environment or field of language models without meaningfully limiting its use on the claims, which does not provide significantly more. Accordingly, even when viewed as a whole, nothing in the claim adds significantly more (i.e. an inventive concept) to the abstract idea. Thus claims 1, 13, and 20 are not patent eligible because the claims are directed to an abstract without significantly more.
Dependent claims 2-9, 12, and 14-19 are also given the full two part analysis, with the additional elements being considered individually and in an ordered combination as a whole, resulting in the following determinations.
Claims 2-6 and 14-18 further limit the abstract idea by reciting steps that merely perform the same process as stated in the independent claim but with additional attributes included such as a commodity code, a product, group, supplier, country and region. Claims 4, and 16 merely send this information to the language model along with the material. Claims 5 and 17 recites performing the step of receiving the word embedding from the language model, which is still part of the abstract idea in the independent claim. Claims 6 and 18 performs the comparing step, but now with the additional data of the attributes factored in. Therefore, these claims recite more of the same abstract idea as the independent claim since they perform the same function but with more data(additional attributes), therefore the claims still recite certain methods of organizing human activity under “commercial or legal interactions.” Furthermore, other than using the same memory, processor, computer, user interface and a non-transitory storage medium to perform more of the same abstract idea, there are no further additional elements to consider in these dependent claims. Therefore, whether considering the additional elements individually, or in an ordered combination, and the claims as a whole including the claims depended on and any intervening claims, claims 2-6 and 14-18 fail to integrate the abstract idea into a practical application or provide significantly more, therefore they are patent ineligible under 35 U.S.C. 101.
Claims 7, 8, and 19 further limit the abstract idea by merely adding the step of filtering the word embeddings to at least have the same commodity code, geography, or validity period. This is more of the same abstract idea because it merely performs the same commercial or legal interactions as the independent claims, but utilizes filtering, which is also part of “certain methods of organizing human activity.” Please see MPEP 2106.04(a)(2)(II)(C). There are no further additional elements to consider, therefore even when considering the claims as a whole, the claim has not been integrated into a practical application and significantly more has not been provided. Therefore claims 7, 8 and 19 are also patent ineligible.
Claim 12 further limits the abstract idea by stating the confidence score is a sum of a cosine similarity score, geography score, a commodity code score, and a temporal score. Since this confidence score is merely part of the emission factor information being outputted, it is more of the same abstract idea of “commercial or legal interactions” because it generally recites what scores are included as part of the confidence score. Even when considering the claims as a whole, including the additional elements previously stated in the independent claim analyzed individually or as a combination, claim 11 is still patent ineligible.
Claim Rejections – 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-9, and 12-20 are rejected under 35 U.S.C. 103 as being rendered obvious Williams et al (US 20250265602 A1, (which has the earliest priority date of 2024-02-15, which is the filing date of provisional application US 63,554,095 which has been attached for reference) hereinafter Williams, in view of Russo et al. (US 12190331 B1) hereinafter Russo.
Regarding Claims 1, 13, 20:
Williams discloses a method of generating carbon footprint/emission factors for physical products, based on minimal user inputs by mapping the product’s materials to known emission factors of historical products. Williams teaches:
Claim 1 Preamble: A system, comprising: at least one processor; and at least one memory including program code which when executed by the at least one processor causes operations comprising: (Williams [0138] The memory (e.g., the main memory 1306, the non-volatile memory 1310, the machine-readable medium 1326) can be local, remote, or distributed. Although shown as a single medium, the machine-readable medium 1326 can include multiple media (e.g., a centralized/distributed database and/or associated caches and servers) that store one or more sets of instructions 1328. [0222] 15. A system for generating a cost estimate and/or an emissions output for a physical product, the system comprising: [0223] a processor; and [0224] a non-transitory computer readable medium storing instructions that, when executed by the processor, cause the processor to)
Claim 13 Preamble: A method comprising: (Williams [0254] 20. A method of generating a carbon footprint for a product to be manufactured at a source location and transported to a destination location, the method comprising:)
Claim 20 Preamble: A non-transitory computer readable store medium including executable code which when executed by at least one processor causes operations comprising: (Williams [0138] The machine-readable medium 1326 can include any medium that is capable of storing, encoding, or carrying a set of instructions for execution by the computer system 1300. The machine-readable medium 1326 can be non-transitory or comprise a non-transitory device. In this context, a non-transitory storage medium can include a device that is tangible, meaning that the device has a concrete physical form, although the device can change its physical state. Thus, for example, non-transitory refers to a device remaining tangible despite this change in state.)
Claim 1 Body (also representative of claims 9 and 10 body):
-receiving, from a user interface, a request including at least one material for which an emission factor suggestion is requested; (Williams [0031] The user application 102 can include a user interface for receiving product inputs 110, displaying carbon emissions outputs 120 associated with the product inputs 110 and generated by the carbon emissions intelligence platform 104, and displaying emissions recommendations 130 generated by the carbon emissions intelligence platform 104... [0032] More specifically, the product inputs 110 can include a product description 111, a declared mass 112, a supplier location 113 (e.g., a source location), a destination location 114 (e.g., a receipt location), a product trade code 115, a procurement cost 116, a supplier name 117, one or more upstream suppliers 118, a product bill of materials (BOM) 119, and/or supplier emissions data 190. [0177] 1. A method of generating a cost estimate and/or an emissions output for a physical product, the method comprising: [0178] receiving inputs specifying the product, wherein the inputs comprise (a) a name of product, (b) a source location where the product is to be sourced, and (c) a destination location where the product is to be received; )
-providing the at least one material to a language model; (Williams [0179] inputting the name of the product into a first artificial intelligence (AI) application and utilizing the first AI application to decompose the product into a plurality of nodes, wherein each node comprises a constituent material and/or component of the product; [0062] At block 591, the method 590 can include utilizing an artificial intelligence (AI) application (e.g., a generative AI application, a generative AI model, a machine learning (ML) model, a large language model (LLM), and/or the like) to convert the given first node to a product trade code, such as an HTS or HS code. For example, the AI application can be a generative AI application in which a text prompt can be input to find the product trade code associated with the first node.)
-in response to the providing, receiving, from the language model, at least one word embedding representative of the at least one material; (Williams [0053] In general, the method 470 can decompose the product into as many levels of nodes (e.g., three, four, or more) as needed until all the nodes are matched to entries in the emissions factors database. In some embodiments, the emissions factors databases are embedded (e.g., vectorized) into the AI application to allow for direct lookups. In some embodiments, the product breakdown databases are embedded (e.g., vectorized) into the AI application to allow for direct lookups. [0134] Product outputs 1220 from the product intelligence platform 1204 can accordingly include a detailed product decomposition into physical nodes and associated key attributes related to cost, emissions, manufacturing, regulatory compliance, etc., with each node. The decomposition, nodes, and attributes can be presented to a user as shown in, for example, FIGS. 8A-10. [0159] An embedding 1406 is a learned numerical representation (such as, for example, a vector) of a token that captures some semantic meaning of the text segment represented by the token 1402. [0160] The vector space can be defined by the dimensions and values of the embedding vectors. Various techniques can be used to convert a token 1402 to an embedding 1406. [0161] The generated embeddings 1406 are input into the encoder 1408. The encoder 1408 serves to encode the embeddings 1406 into feature vectors 1414 that represent the latent features of the embeddings 1406.) In [0053] William’s explains that the products are embedded into the AI application, and in [0134] product outputs include decompositions, nodes, and attributes. Since William’s decompositions, nodes and attributes can include word embeddings which are taught in [0159-161], then these nodes satisfy the limitation of “word embeddings representative of at least one material. Note that Williams teaches in [0062] that the “AI Application” includes large language models (LLM).
-initiating, via an application programming interface (API), a semantic search of a vector database...(Williams [0030] In some embodiments, the user application 102 is a cross-platform software application configured to work on several computing platforms and web browsers and/or an application programming interface (API). [0167] Inputs to an LLM can be referred to as a prompt, which is a natural language input that includes instructions to the LLM to generate a desired output. A computer system can generate a prompt that is provided as input to the LLM via an API. [0106] Some or all of the emissions indicators 848-852 can be interactive. That is a user can interact with the user interface 840 via a user input (e.g., click of a mouse, positioning of a cursor, and/or the like) to reveal different and/or additional information. [0171] An application 1520 interfaces between a user or external system and the architecture of the LLM. A query 1522 can be input at the application 1520. Based on the query, the application 1520 generates a prompt or series of prompts to cause the LLM to produce a specified output. The application 1520 returns outputs 1524 from the LLM to the requesting user or system. [0050] If the product does not match an entry in the emissions factors databases, the method 470 can proceed to block 473 to utilize an artificial intelligence (AI) application (...a retrieval-augmented generation (RAG) model, and/or the like) to decompose the product into its components, materials, and/or manufacturing processes, which can be referred to as first nodes. For example, the AI application can be a generative AI application in which a text prompt can be input to decompose the product into its components, materials, and/or manufacturing processes, and/or the AI application can be a RAG application in which information about the product breakdown is retrieved from a product breakdown database (e.g., the product breakdown database 145 of FIG. 1). [0079] retrieval-augmented generation (RAG) model, and/or the like) to find the most relevant activities in the emissions factor databases or manufacturing emissions databases that match to the manufacturing steps generated by the manufacturing process generator 266. For example, the AI application can be a RAG application in which the manufacturing step information can be input to find the closest matching manufacturing emission factor activity in the manufacturing emissions factor databases. [0159] An embedding 1406 is a learned numerical representation (such as, for example, a vector) of a token that captures some semantic meaning of the text segment represented by the token 1402. The embedding 1406 represents the text segment corresponding to the token 1402 in a way such that embeddings corresponding to semantically related text are closer to each other in a vector space than embeddings corresponding to semantically unrelated text.) Since LLM invokes the other steps of the embodiments such as the vector database search and comparison, then it can be said that the query input at the application “initiates a search of a vector database.” Williams satisfies the “semantic search” limitation because “retrieval-augmented generation” (in [0050] and [0079]) is known to a person of ordinary skill in the art to fall within the umbrella of “semantic search,” and furthermore, William’s teachings that the vectors hold semantic meaning ([0159]), therefore, a lookup of the vectors satisfies “semantic search.”
-database storing a plurality of word embeddings representative of a plurality of materials each mapped to a corresponding emission factor stored in an emission factors database (Williams [0132] For example, for low level material nodes, the emissions mapper 1254 can determine the most applicable material/component emission factor in an emissions factor database. This can be a direct material-to-material match or, depending on coverage, a relevant proxy may be used. Proxy matching uses vector similarity scores, determining the best matching material/component in the vector database. [0170] The data preprocessing block 1510 can generate embeddings of the contextual data or invoke a service to generate the embeddings. The models used to generate embeddings can be trained for the specific model or application in which the embeddings are to be used. Embeddings can be stored in a vector database. [0074] For example, the node can previously be matched to an entry in the emissions factors databases via the method 470 described in detail with reference to FIG. 4. More specifically, for example, referring to FIG. 3B, the first node 380b “dyed cotton” matches to the primary entry 382 “dyeing of cotton yarn” in the emissions factors databases, and the second node 384 matches to the primary entry 385 “plastic” in the emissions factors databases. [0079] For example, the AI application can be a RAG application in which the manufacturing step information can be input to find the closest matching manufacturing emission factor activity in the manufacturing emissions factor databases. [0053] In some embodiments, the emissions factors databases are embedded (e.g., vectorized) into the AI application to allow for direct lookups. In some embodiments, the product breakdown databases are embedded (e.g., vectorized) into the AI application to allow for direct lookups.) “Lookups,” “Retrieval Augmented Generation,” and “Find the closest matching manufacturing emission factor activity,” all fall within the scope of “searching a database storing a plurality of word embeddings.” At least [0132, 0170, 0074, 0079, 0053] are shown to satisfy the “database storing a plurality of word embeddings each mapped to a corresponding emission factor stored in an emission factors database,” but more evidence supporting William’s database satisfying the limitation can be found in at least [0049-0052], [0054-0061], [0036] (find “corresponding emissions factors”) .
- wherein the search comprises comparing, in a vector space defined by the plurality of word embeddings, the at least one word embedding representative of the at least one material to at least a portion of the plurality of word embeddings; (Williams [0048] The product decomposer 257 can utilize the product breakdown database 145 (FIG. 1) to search for products that are similar to the input product, subsequently using the retrieved information for the product breakdown. [0050] For example, the AI application can be a generative AI application in which a text prompt can be input to decompose the product into its components, materials, and/or manufacturing processes...At block 474, for each first node, the method 470 can determine whether the first node matches an entry in the emissions factors database (e.g., the material/component emissions factors database 140 and/or the manufacturing emissions factors database 141 of FIG. 1). If the first node matches an entry in the emissions factors database, the method 470 can proceed to block 472 to end further processing for that first node. [0035] As a specific example, the emissions factors databases 140 can store an emissions factor (e.g., a quantity of carbon dioxide released per unit mass (e.g., 1 kg)) for polyester resin, energy inputs for producing polyester resin, and/or process emissions output from the production process. [0057] For example, the first node 380a “buttons” is decomposed into a single second node 384 “plastic.” At block 476, the second node 384 matches an entry in the emissions factors databases. Specifically, the second node 384 matches to a primary entry 385 “plastic” in the emissions factors database. The primary entry 385 can be a composite of multiple separate entries in the ecoinvent and/or other emissions databases as described in detail above. ) Since Williams compares the text input to see if the nodes match an entry in the emissions factor database(which includes word embeddings), then Williams satisfies the limitations because it compares word embeddings representative of a material to the plurality of stored word embeddings. As seen in paragraph [0035], the emissions factor database includes the materials and their associated emissions factors. See also [0132] below which further describes the search and mapping process.
-identifying, based on a similarity metric, at least one matching word embedding for the at least one word embedding representative of the at least one material, wherein the at least one matching word embedding maps to at least one emission factor stored in the emissions factor database; and (Williams [0132] The emissions mapper 1254 can receive the decomposed nodes, manufacturing steps, and transport estimates, and determined estimated emissions of the product based on each as, for example, described in detail above with reference to the product emissions estimator 258 of FIG. 2. For example, for low level material nodes, the emissions mapper 1254 can determine the most applicable material/component emission factor in an emissions factor database. This can be a direct material-to-material match or, depending on coverage, a relevant proxy may be used. Proxy matching uses vector similarity scores, determining the best matching material/component in the vector database. For manufacturing nodes, the emissions mapper 1254 can determine the most applicable manufacturing emission factors for each of the manufacturing steps determined by the manufacturing modeler 1251. This can use a similar direct or proxy mapping as the previous case. [0053] In general, the method 470 can decompose the product into as many levels of nodes (e.g., three, four, or more) as needed until all the nodes are matched to entries in the emissions factors database. In some embodiments, the emissions factors databases are embedded (e.g., vectorized) into the AI application to allow for direct lookups. In some embodiments, the product breakdown databases are embedded (e.g., vectorized) into the AI application to allow for direct lookups. [0058] The emissions factors for nodes with direct connections to entries in the emissions factors database (e.g., the first node 380b and the second node 384) can be retrieved/extracted from the matching entries in the emissions factors databases (e.g., the primary entry 382 for the first node 380b and the primary entry 385 for the second node 384). The emissions factors for nodes without direct connections to entries in the emissions factors database (e.g., the primary node 389 and the first node 380a) cannot be retrieved/extracted from the emissions factors database in the same manner. [0159] An embedding 1406 is a learned numerical representation (such as, for example, a vector) of a token that captures some semantic meaning of the text segment represented by the token 1402. The embedding 1406 represents the text segment corresponding to the token 1402 in a way such that embeddings corresponding to semantically related text are closer to each other in a vector space than embeddings corresponding to semantically unrelated text. For example, assuming that the words “write,” “a,” and “summary” each correspond to, respectively, a “write” token, an “a” token, and a “summary” token when tokenized, the embedding 1406 corresponding to the “write” token will be closer to another embedding corresponding to the “jot down” token in the vector space as compared to the distance between the embedding 1406 corresponding to the “write” token and another embedding corresponding to the “summary” token. [0160] The vector space can be defined by the dimensions and values of the embedding vectors.) William’s vector similarity scores is an example of a similarity metric used to identify the matching word embedding(material-to-material match).
-sending, to the user interface, a response including the at least one emission factor and a corresponding confidence score based on the similarity metric to indicate a similarity between the at least one word embedding and the at least one matching word embedding. (Williams [0094] Referring again to FIG. 2, an uncertainty estimator 264 of the carbon emissions intelligence platform 104 can calculate an uncertainty (mapped to confidence score) for the emissions factors estimate for each node and the various transport steps. [0095] The output aggregator 265 can output the emissions outputs 120 for display at, for example, the user application 102 (FIG. 1). The output aggregator 265 can further aggregate the uncertainties generated by the uncertainty estimator 264 to generate an aggregate uncertainty (and/or confidence interval) associated with the emissions outputs 120. In some embodiments, the uncertainty values can be aggregated by a statistical metric (mapped to similarity metric) based on other values of the nodes. For example, a weighted average of uncertainty can be aggregated with weight being based on the emissions of the node. [0080] In some embodiments, the AI application can return multiple closely matching activities from the manufacturing emissions factor databases and take a statistical approach to average the emission factors. For example, a weighted average based on likelihood of match(mapped to similarity metric to indicate a similarity between the embedding and matching embedding) can be used to weight the final emission factor based on the likelihood of the activities being a good match to the manufacturing step. Such a likelihood can be provided as part of the AI application output.) In Williams, the output aggregator outputs the emissions outputs for display on the user application, including the uncertainties calculated. William’s teaches that these uncertainty values can be aggregated by a statistical metric, which is mapped to similarity metric. In [0080], Williams shows a similarity metric that indicates the similarity between the two matching word embeddings (in the form of a likelihood of match).
-storing the at least one word embedding representative of the at least one material in the vector database...maps the at least one word embedding to the at least one emission factor stored in the emission factors database.(Williams [0035] The emissions factors databases 140 can store a dataset of emissions factors, emissions associated with energy inputs (e.g., electricity, fuel, etc.), emissions associated with process outputs (e.g., from chemical reactions, etc.), and/or the like for different materials and components that may be used in a product. An emissions factor is a representative value that relates the quantity of carbon dioxide or other pollutant released to the atmosphere with an activity associated with the different materials and components. As a specific example, the emissions factors databases 140 can store an emissions factor (e.g., a quantity of carbon dioxide released per unit mass (e.g., 1 kg)) for polyester resin, energy inputs for producing polyester resin, and/or process emissions output from the production process. The emissions factor can aggregate emissions generated from the production of the raw materials used to produce the polyester resin (e.g., acetic anhydride, adipic acid, etc.), from the processes used to transform the raw materials into polyester resin, etc. [0042] The databases 106 can be periodically updated (e.g., annually, biyearly, monthly, or more frequently) with new data. In some embodiments, the carbon emissions intelligence platform 104 can directly provide data from one or more of the databases 140 related to the product inputs to the user application 102. For example, data from the material/component emissions factors databases 140 can be provided to the user application 102. [0053] In general, the method 470 can decompose the product into as many levels of nodes (e.g., three, four, or more) as needed until all the nodes are matched to entries in the emissions factors database. In some embodiments, the emissions factors databases are embedded (e.g., vectorized) into the AI application to allow for direct lookups. In some embodiments, the product breakdown databases are embedded (e.g., vectorized) into the AI application to allow for direct lookups. [0132] The emissions mapper 1254 can receive the decomposed nodes, manufacturing steps, and transport estimates, and determined estimated emissions of the product based on each as, for example, described in detail above with reference to the product emissions estimator 258 of FIG. 2. For example, for low level material nodes, the emissions mapper 1254 can determine the most applicable material/component emission factor in an emissions factor database. This can be a direct material-to-material match or, depending on coverage, a relevant proxy may be used. Proxy matching uses vector similarity scores, determining the best matching material/component in the vector database. [0170] The data preprocessing block 1510 can generate embeddings of the contextual data or invoke a service to generate the embeddings. The models used to generate embeddings can be trained for the specific model or application in which the embeddings are to be used. Embeddings can be stored in a vector database.)
However, Williams fails to teach:
-that the at least one word embedding representative of the at least one material stored in the vector database is stored “with a key” that maps the at least one word embedding to the at least one emission factor stored in the emissions factor database. (We know from Williams [0035], [0053], and [0132] that Williams teaches the functions of storing in a manner of maps that the word embeddings to the emission factor in the emissions factor database. However, Williams does not explicitly teach that this mapping is “a key.”)
Alternatively, Russo discloses a method of carbon emission optimization using machine-learning which involves vector embedding techniques such as a word2vec, and stores the integrated logistic data with an associated emission factor in a relational database. Russo teaches:
-storing the at least one word embedding in the vector database with a key that maps the at least one word embedding to the at least one emission factor stored in the emission factors database.(Russo [Col. 13 Line 65 – Col. 14 Line 21] Still referring to FIG. 1, to transform integrated logistics data collection 112 into logistic vector space, various vector embedding techniques may be employed, by processor 104. As used in this disclosure, “vector embedding techniques” are algorithms or methods that convert raw data into vectors in a way that preserves the inherent relationships and structures in the data. Exemplary vector embedding techniques may include, without limitation, Word2Vec,..., Word2Vec may be adapted to represent logistics data by treating each data point or datasets within integrated logistics data collection 112 as a semantic unit e.g., a “word”...” In a non-limiting example, if integrated logistics data collection includes transportation data such as routes, vehicle types, and schedules, wherein each route or vehicle type may be treated as a word or a set of words. Using Word2Vec, each of these “words” may be embedded into logistic vector space, resulting in vectors that capture the relationships between different routes or vehicle types based on their co-occurrence in integrated logistics data collection 112. [Col. 10 Lines 25-47] identifying an emission factor associated with integrated logistic data collection 112. As used in this disclosure, an “emission factor” refers to a representative value that attempts to relate the quantity of a pollutant as described herein release to the atmosphere with certain data points (e.g., events) associated with the release of the pollutant...emission factor may be associated with one or more data points or datasets such as item/product type, transportation mode, fuel type, or even operational practices directly or indirectly specified in integrated logistic data collection 112. [Col. 7 Lines 9-31] With continued reference to FIG. 1, in an embodiment, processor 104 may receive integrated logistic data collection 112 and/or any data described herein from a logistic database 128, wherein the logistic database 128 may be implemented, without limitation, as a relational database, a key-value retrieval database such as a NOSQL database... Logistic database 128 may include a plurality of data entries and/or records as described above. Data entries in a database may be flagged with or linked to one or more additional elements of information, which may be reflected in data entry cells and/or in linked tables such as tables related by one or more indices in a relational database. Persons skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various ways in which data entries in a database may store, retrieve, organize, and/or reflect data and/or records as used herein, as well as categories and/or populations of data consistently with this disclosure. [Col. 8 Lines 39-57] referring to FIG. 1, receiving integrated logistic data collection 112 may include aggregating data such as, without limitation, order data 116, transportation data 120, environmental data 124, or any other data described herein using data fusion techniques... In another non-limiting example, processor 104 may be configured to join one or more data structures. Tables in logistic databases 128 as described above based on pre-defined relations such as common keys (e.g., primary, and foreign keys);) Therefore, since Russo teaches storing logistics data (which includes material/item information) as word embeddings in a vector database in [Col. 13 65 – Col. 14 21], and teaches emission factors associated with the logistic data in [Col. 10 Lines 25-47], and also teaches that the logistic data tables can be related using key-value retrieval, then it is clear that Russo satisfies the limitation above in [Col. 8 Lines 39-57] because the logistic data (including the embeddings and related emission factor) are stored with primary and foreign keys that map the embedding to the emission factor.
Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the present disclosure to modify Williams by using relational databases as taught by Russo to store a key along with the embedding information. By simply adding a primary key or foreign key to Williams which relates the materials database to an emissions factor database as taught by Russo, one would reasonably expect to arrive the claimed limitation. One of ordinary skill in the art would have been motivated to perform this combination as it would expand the ability of the databases to be related to existing data such as data from public databases as taught by Russo, which would have been beneficial to William’s system which also sourced and scraped public data. (Russo [Col. 10 Lines 41-46] In some cases, processor 104 may retrieve pre-existing emission factors from logistic database 128 or any other public databases affiliated with environmental agencies or international organizations and contain emission factors for a wide range of activities and conditions.)
Regarding Claims 2, 14:
The combination of Williams and Russo teach or suggest The system of claim 1/ The method of claim 13
Furthermore, Williams teaches:
-wherein the request may be received with one or more additional attributes associated with the at least one material. (Williams [0032] More specifically, the product inputs 110 can include a product description 111, a declared mass 112, a supplier location 113 (e.g., a source location), a destination location 114 (e.g., a receipt location), a product trade code 115, a procurement cost 116, a supplier name 117, one or more upstream suppliers 118, a product bill of materials (BOM) 119, and/or supplier emissions data 190. The product description 111 can comprise text, images, and/or other information specifying a product (e.g., the text “t-shirt,” “Henley t-shirt,” “circuit board,” “athletic shoe,” “alkaline battery,” an image of a-shirt, an image of a Henley t-shirt, an image of a circuit board, an image of an athletic shoe, an image of an alkaline battery, etc.).) All of the inputs regarding the product, other than just the name of the product, are all additional attributes associated with the material.
Regarding Claims 3, 15:
The combination of Williams and Russo teach or suggest The system of claim 2/ The method of claim 14
Furthermore, Williams teaches:
- wherein the additional attributes comprise a commodity code, (Williams [0032] product trade code [0063] In some embodiments, the AI application can match the first node to a product trade code having a maximum number of digits (e.g., six) and corresponding specificity while, in other instances, the AI application may only be able to match the first node to a product trade code having a fewer number of digits (e.g., two, four) and corresponding lesser specificity.)
-a product group, (Williams [0063] For example, referring to FIG. 3B, the HS code for the first node 380a, a button, is 96.06, where the first two digits “96” designate the chapter (e.g., chapter 96: “Miscellaneous Manufactured Articles”) and the final two digits “06” designate the heading within the chapter (e.g., “buttons”). In some embodiments, the AI utilizes information from lower-tier nodes, such as the second node 384, to further specify the product trade code.)
-a supplier, (Williams [0066] In some embodiments, the product inputs 110 (FIGS. 1 and 3A) can optionally include the upstream suppliers 118 (FIGS. 1 and 3A) that specify one or more known upstream suppliers. In such instances, at block 592, the method 590 can only determine total imports of the product trade code from the countries of the specified upstream suppliers 118.)
-a country, and/or a region. (Williams [0065] Referring again to FIG. 5, at block 592, the method 590 can include determining the total imports of the product trade code (e.g., ranging from 2-6 digits) to a source country (e.g., the location of the supplier specified by the supplier location 113 of the product inputs 110 of FIG. 1) from countries foreign to the source country based on one or more trade modeling databases (e.g., the trade modeling databases 142).)
Regarding Claims 4, 16:
The combination of Williams and Russo teach or suggest The system of claim 2/ The method of claim 14
Furthermore, Williams teaches:
- wherein the providing further comprises providing the at least one material and the one or more additional attributes to the language model. (Williams [0121] Referring to FIGS. 1, 2, and 10, in some embodiments the products emissions estimator 258 and/or another component of the platform 100 can use an artificial intelligence (AI) application (e.g., a generative AI application, a generative AI model, a machine learning (ML) model, a large language model (LLM), and/or the like) to convert/map the first nodes (e.g., identified at second graph portions 1065), the second nodes (e.g., identified at third graph portions 1066), and/or further nodes of the product decomposition to a product trade code, such as an HTS or HS code. The databases 106 can further comprise one or more pricing databases that contain information about commodity market prices of components, materials, labor prices, and/or energy prices, etc., based on industry classification code or component/material name and/or description.)
Regarding Claims 5, 17:
The combination of Williams and Russo teach or suggest The system of claim 4/ The method of claim 16
Furthermore, Williams teaches:
- wherein in response to the providing, the receiving, from the language model, further comprises receiving at least one word embedding representative of the at least one material and the one or more additional attributes. (Williams [0075] a large language model (LLM), and/or the like) to convert the node to an industry classification code, such as a North American Industry Classification System (NAICS) code. For example, the AI application can be a generative AI application in which a text prompt can be input to find the industry classification code associated with the node. [0079] a large language model (LLM), retrieval-augmented generation (RAG) model, and/or the like) to find the most relevant activities in the emissions factor databases or manufacturing emissions databases that match to the manufacturing steps generated by the manufacturing process generator 266. [0085] a large language model (LLM), and/or the like) to determine the most likely modes (e.g., sequence of one or more modes) of transport from a supplier location to a destination location for the given product.) As seen above, the language models are also capable of receiving the nodes and outputting the data along with the additional attributes. See also [0132].
Regarding Claims 6, 18:
The combination of Williams and Russo teach or suggest The system of claim 5/ The method of claim 17
Furthermore, Williams teaches:
- wherein the comparing comprises comparing the at least one word embedding representative of the at least one material and the one or more additional attributes to the plurality of word embeddings. (Williams [0057] In the illustrated embodiment, at block 474 the first node 380a does not match an entry in the emissions factors databases such that the method 470 proceeds to block 475 to utilize the AI application to decompose the first node 380a into one or more second nodes 384 of components, materials, and/or manufacturing processes. For example, the first node 380a “buttons” is decomposed into a single second node 384 “plastic.” At block 476, the second node 384 matches an entry in the emissions factors databases. Specifically, the second node 384 matches to a primary entry 385 “plastic” in the emissions factors database. The primary entry 385 can be a composite of multiple separate entries in the ecoinvent and/or other emissions databases as described in detail above...More specifically, the emissions factor associated with the primary entry 385 can specify the energy inputs (e.g., electricity, fuel, etc.) for the second node 384 and their corresponding carbon (and/or CO.sub.2e) emissions and the process emissions (e.g., from chemical reactions, etc.) for the second node 384. Accordingly, after identifying the match to the primary entry 385 in the emissions databases, the method 470 can proceed to block 472 to end processing for the second node 384.) In this excerpt of Williams, Williams teaches that the matching of nodes to entries in emission factors databases, don’t only match the word embeddings but also additional attributes such as manufacturing processes and energy inputs.
Regarding Claims 7, 19:
The combination of Williams and Russo teach or suggest The system of claim 1/ The method of claim 13
Furthermore, Williams teaches:
- wherein the semantic search further comprises limiting the searching to only the plurality of word embeddings having a same commodity code as the at least one material. (Williams [0063] In some embodiments, the AI application can match the first node to a product trade code having a maximum number of digits (e.g., six) and corresponding specificity while, in other instances, the AI application may only be able to match the first node to a product trade code having a fewer number of digits (e.g., two, four) and corresponding lesser specificity. For example, referring to FIG. 3B, the HS code for the first node 380a, a button, is 96.06, where the first two digits “96” designate the chapter (e.g., chapter 96: “Miscellaneous Manufactured Articles”) and the final two digits “06” designate the heading within the chapter (e.g., “buttons”). [0064] The product inputs 110 (FIGS. 1 and 3A) can optionally include the product trade code 115 (FIGS. 1 and 3A) that specifies one or more product trade codes for the product and/or its components and materials. In such instances, the AI application can utilize the product trade code 115 to narrow, focus, or specify the conversion of the first node to a product trade code.) The product trade code can be used to “narrow, focus, or specify...” which means that the searching is limited to the matching commodity codes.
Regarding Claim 8:
The combination of Williams and Russo teach or suggest The system of claim 1
Furthermore, Williams teaches:
- wherein the identifying further comprises filtering the at least one matching word embedding based on geography. (Williams [0043] The carbon emissions outputs 120 can comprise total carbon emissions 121, an emissions factor 122, a comparison to global 123, an emission by country 124, [0065] For example, the AI application can query the United Nations ComTrade database and/or the CEPII BACI database to determine the total imports of the product trade code to the source country on a per-country basis. [0126] The regulatory compliance databases can comprise a classification of materials by country or based on country (e.g., a material sourced from China may be classified differently than a material sourced from another country). The AI application can query the regulatory compliance databases to generate a per-node (e.g., per-component or per-material) flagging for regulatory compliance of the product (e.g., the inertial measurement unit).)
Regarding Claim 9:
The combination of Williams and Russo teaches or suggests the system of claim 1
Furthermore Williams teaches:
- wherein the similarity metric comprises a similarity metric determined between the at least one word embedding representative of the at least one material and the plurality of word embeddings. (Williams [0132] The emissions mapper 1254 can receive the decomposed nodes, manufacturing steps, and transport estimates, and determined estimated emissions of the product based on each as, for example, described in detail above with reference to the product emissions estimator 258 of FIG. 2. For example, for low level material nodes, the emissions mapper 1254 can determine the most applicable material/component emission factor in an emissions factor database. This can be a direct material-to-material match or, depending on coverage, a relevant proxy may be used. Proxy matching uses vector similarity scores, determining the best matching material/component in the vector database.)
However, Williams does not teach:
-the similarity metric comprises a cosine similarity
Alternatively, Russo teaches or suggests:
-the similarity metric comprises a cosine similarity(Russo [Col. 8 Lines 13-17] Vectors may be more similar where their directions are more similar, and more different where their directions are more divergent, for instance as measured using cosine similarity as computed using a dot product of two vectors; [Col. 14 Lines 43-60] In some cases, vector similarity may be measured according to any norm for proximity and/or similarity of two vectors, including without limitation cosine similarity. As used in this disclosure “cosine similarity” is a measure of similarity between two-non-zero vectors of a vector space, wherein determining the similarity includes determining the cosine of the angle between the two vectors. Cosine similarity may be computed as a function of using a dot product of the two vectors divided by the lengths of the two vectors, or the dot product of two normalized vectors. For instance, and without limitation, a cosine of 0° is 1, wherein it is less than 1 for any angle in the interval (0,π) radians. Cosine similarity may be a judgment of orientation and not magnitude, wherein two vectors with the same orientation have a cosine similarity of 1, two vectors oriented at 90° relative to each other have a similarity of 0, and two vectors diametrically opposed have a similarity of −1, independent of their magnitude.)
Therefore, it would have been obvious to one of ordinary skill in the art to modify Williams by teaching the similarity metric specifically uses a “cosine similarity” as taught by Russo. It would have been obvious to simply substitute William’s vector similarity with cosine similarity, as it is known that cosine similarity is one of the main methods to calculate vector similarity. One would have been motivated to perform this combination by the fact that cosine similarity determines whether vectors are more similar in direction and more different when their directions are divergent. (Russo [Col. 8 Lines 7-23] Still referring to FIG. 1, two vectors may be considered equivalent where their directions, and/or the relative quantities of values within each vector as compared to each other, are the same; thus, as a non-limiting example, a second order vector [10.0, 0.6, 4.8] may be treated as equivalent, for purposes of this disclosure, as order vector v.sub.order as described above. Vectors may be more similar where their directions are more similar, and more different where their directions are more divergent, for instance as measured using cosine similarity as computed using a dot product of two vectors; however, vector similarity may alternatively or additionally be determined using averages of similarities between like attributes, or any other measure of similarity suitable for any n-tuple of values, or aggregation of numerical similarity measures for the purposes of loss functions as described in further detail below.])
Regarding Claim 12:
Williams teaches the system of claim 1
Furthermore Williams teaches:
- wherein the confidence score comprises a sum of a (vector) similarity score, a geography score, (Williams [0095] The output aggregator 265 can further aggregate the uncertainties generated by the uncertainty estimator 264 to generate an aggregate uncertainty (and/or confidence interval) associated with the emissions outputs 120. In some embodiments, the uncertainty values can be aggregated by a statistical metric based on other values of the nodes. For example, a weighted average of uncertainty can be aggregated with weight being based on the emissions of the node. [0132] Proxy matching uses vector similarity scores, determining the best matching material/component in the vector database...For all matched emission factors, the emissions mapper 1254 can adjust the emission factor to the local grid or most likely fuel type... As a specific example, a material node of chlorine may be directly matched to a chlorine production activity in the base database, then given the manufacturing is occurring in Mexico the electricity and fuel can be adjusted with the Mexico grid emission factor and most likely fuel type ratio (natural gas, coal, etc.) for the location, respectively. ) The broadest reasonable interpretation of this limitation is that the confidence score is calculated at least in part by summing a vector similarity score, geography score, and commodity code score, without specifically defining how each of these terms are calculated. Therefore, since the aggregate uncertainties are based on a statistical metrics based on the other values of the nodes in Williams, and William’s includes metrics in nodes such as vector similarity score, Mexico grid emission factor (mapped to geography score), and
- a commodity code score.(Williams [0077] Referring again to FIG. 6A, at block 604 the AI application can determine whether the industry classification code for the node matches an entry in the emissions factors databases, such as the manufacturing emissions factors databases 141 of FIG. 1. In some embodiments, the AI application finds the best match for the industry classification code by first starting with a 6-digit code (if determined), then moving to more general 5-digit, 4-digit, 3-digit, and 2-digit codes if the more specific code does not match an entry in the manufacturing emissions factors databases 141...a given industry classification code may match multiple entries in the manufacturing emissions factors database. In such instances, the method 600 can return the average, median, or other weighted blend of the emissions factors matching the industry classification code. In some embodiments, the method 600 can randomly sample the multiple entries using, for example, a Monte Carlo analysis, to determine an uncertainty associated with the calculated emissions factors. If no, the method 600 can proceed to block 605 to return an error message.) In Williams, the uncertainty is also calculated based on the degree of similarity to the classification codes.
-temporal score (Williams [0101] FIG. 8B illustrates the display of an emissions report on the user interface 840 for the inputs entered into the input field shown in FIG. 8A (and as described in detail above with reference to FIGS. 2-7). In the illustrated embodiment, the emissions report includes a total emissions value 845 for the cradle-to-gate life cycle (e.g., from extraction of raw materials, through manufacturing, and through shipping to the product destination in Seattle, WA) of the Henley t-shirt, an emissions factor value 846 for the cradle-to-gate life cycle of the Henley t-shirt, and a comparison to global value 847 for the cradle-to-gate life cycle of the Henley t-shirt. [0106] Likewise, in some embodiments positioning the cursor over and/or clicking on any of the flow paths of the contribution analysis 851 reveals the total emissions and/or emissions factor for that portion of the cradle-to-gate life cycle of the Henley t-shirt.) The broadest reasonable interpretation is any metric that accounts for the validity period of the emissions factor in view of [0058]. Since William’s accounts for the “cradle-to-gate life cycle”, which is a time period, the claims as broadly stated are satisfied.
However, Williams does not teach:
-the confidence score comprises a sum of cosine similarity score,
Alternatively, Russo teaches:
-the confidence score comprises a sum of cosine similarity score, (Russo [Col. 15 Lines 1-19] Still referring to FIG. 1, in some cases, dimensions of logistic vector space may not represent distinct data within integrated logistics data collection 112, in which case elements of a vector representing a first transportation data may have numerical values that together represent a geometrical relationship to a vector representing a second transportation data, wherein the geometrical relationship represents and/or approximates a relationship between the first transportation data and the second transportation data. For instance, and without limitation, an angle between two logistics vectors may capture the correlation between two transportation datasets they represent. In a non-limiting example, if the cosine similarity between two logistics vectors is close to 1, it may indicate that two corresponding transportation datasets have very similar logistics patterns or attributes. On the other hand, a cosine similarity close to −1 may suggest that two corresponding transportation datasets are diametrically opposite in one or more logistics aspects described herein.) The cosine similarity between two logistics vectors is inherently “a sum of cosine similarity” score.
Therefore, it would have been obvious to one of ordinary skill in the art to modify Williams by teaching the confidence score also includes a sum of “cosine similarity” as taught by Russo. It would have been obvious to simply substitute William’s vector similarity with cosine similarity, as it is known that cosine similarity is one of the main methods to calculate vector similarity. One would have been motivated to perform this combination by the fact that cosine similarity determines whether vectors are more similar in direction and more different when their directions are divergent. (Russo [Col. 8 Lines 7-23] Still referring to FIG. 1, two vectors may be considered equivalent where their directions, and/or the relative quantities of values within each vector as compared to each other, are the same; thus, as a non-limiting example, a second order vector [10.0, 0.6, 4.8] may be treated as equivalent, for purposes of this disclosure, as order vector v.sub.order as described above. Vectors may be more similar where their directions are more similar, and more different where their directions are more divergent, for instance as measured using cosine similarity as computed using a dot product of two vectors; however, vector similarity may alternatively or additionally be determined using averages of similarities between like attributes, or any other measure of similarity suitable for any n-tuple of values, or aggregation of numerical similarity measures for the purposes of loss functions as described in further detail below.])
Response to Arguments
The applicant’s remarks filed on 12/03/2025 have been fully considered but are not persuasive for the following reasons.
Regarding arguments over claim rejections under 35 U.S.C. 101, the applicant argues in support of the assertion that the additional elements integrate the abstract idea into a practical application as “the claim improves the technical field of emission factor mapping for enterprise resource planning (ERP) systems.” However, the examiner respectfully disagrees. The examiner notes that emission factor mapping is not necessarily a technical field as it is part of a business practice under “certain methods of organizing human activity.” Secondly, the scope of the claims is not limited to “enterprise resource planning (ERP) systems,” nor do they reflect an improvement in enterprise resource planning systems. The applicant further argues based on the specification that the claim supports purported improvements in [0015] and [0016] of the instant specification by providing suggestions for the materials based on a semantic search. However, the examiner notes that the purported improvement is more in line with an improvement to the abstract idea instead of a technical improvement. MPEP 2106.05(a) specifically states, “Notably, the court did not distinguish between the types of technology when determining the invention improved technology. However, it is important to keep in mind that an improvement in the abstract idea itself (e.g. a recited fundamental economic concept) is not an improvement in technology.” Therefore, improving suggestions for materials based on emission factor data alone is not an improvement in technology, especially when the improvement is lent by generally linking the use of generic natural language processing. Improving an abstract idea by “applying” a general purpose computer or a technological field of use does not satisfy the improvement to technology requirement in MPEP 2106.05(a). The improvement must be an improvement to the computer functionality or field of use itself (in this case an improved natural language processing method). Therefore, the applicant’s arguments regarding an alleged technical improvement are not persuasive.
In response to the applicant’s assertion that the improvement is reflected in the claimed invention in the “initiating,” “comparing,” and “storing” steps, the examiner respectfully disagrees. The provided steps are still directed to the abstract idea because they are recited at a high level of generality and merely claim the idea of the outcome without a specific set of steps or mechanisms to arrive at the outcome. In other words, “initiating a semantic search” on a vector database storing a plurality of word embeddings, wherein the “semantic search” comprises comparing the word embedding of the material to a portion of word embeddings, under its “broadest reasonable interpretation” is not recited at a level of specificity that meaningfully limits the abstract idea into significantly more. At this level of generality, the steps still encompass mere instructions to an individual to carry out any way of performing the outcome, and merely provide generic data collection, analysis, and storage steps that are not exclusive to a specific technological implementation. In other words, “storing the embedding representative of at least one material with a key that maps the at least one word embedding to the at least one emission factor” can be performed by a human with a pen and paper, wherein the key is merely an identifier number, the embedding is merely a 2 or 3 dimensional series of numbers reflecting a material, and the comparison encompasses any comparison of vectors. The fact that the search is initiated on an “application programming interface (API)” is the only additional element that merely instructs the abstract idea to be performed on a computer, however, it does not integrate the abstract idea into a practical application or provide significantly more because it is equivalent to “apply it.” The applicant’s arguments that the steps are inherently technical concepts that require a data processor for implementation are thereby unpersuasive because other than mere instructions to “apply it” on a generic computer, the functions themselves can be practically performed in the human mind with a pen and paper. The claims do not recite any specific steps at a least of complexity that is beyond what is capable in the human mind, because even when considering “cosine similarity,” this equation has been commonplace in linear algebra for ages.
Furthermore, the applicant’s arguments in page 10 which state that the “semantic search of a vector database” enables automatic retrieval of emission data for materials not previously evaluated by the system. This argument is not persuasive because the scale of the vector database and the storage of a word embedding as reflected in the scope of the claim, though it could be performed faster inherently in a computer, is not necessarily too large or complex for the human mind. Secondly, the applicant’s argument that “the quality of results provided to the users of the ERP system are continually improved” is not persuasive because improving the quality of outputs to an individual still falls within the scope of an abstract idea, whether they are on an outputted on an ERP system or not. Secondly, storing an embedding with a key does not reflect a technological improvement nor does it improve the field of computer storage.
In response to the applicant’s citation of the August 4, 2025 memo, the examiner acknowledges the memo, affirms that the claims have been considered as a whole, including the additional elements individually or in combination, and stands by the rejection of the claims under 35 U.S.C. 101.
In response to applicant’s arguments over the prior art under 35 U.S.C. 102(a)(2), the applicant’s remarks have been fully considered but are either moot in view of the updated combination under 35 U.S.C. of Williams in view of Russo, or would not have been persuasive. Firstly, the applicant asserts that “Williams fails to teach a semantic search of a vector database by comparing, in a vector space defined by a plurality of word embeddings, at least one word embedding representative of at least one material to at least a portion of the plurality of word embeddings.” However, the examiner respectfully disagrees. Though Williams does not use the wording “semantic search,” Williams process’ falls within the scope of the claimed limitation as William initiates a retrieval augmented generation process [0050] (which inherently includes a semantic search), because the embeddings are taught by Williams to hold semantic data[0159]. The applicant’s argument that comparing a node against entries..., is completely different than comparing a word embedding against another word embedding, is not persuasive because Williams teachings comparing embeddings in a vector space on numerous occasions. [0035][0053][0132]. While some embodiments in Williams are broad enough to teach “comparing a node against entries in an emissions factor database,” it is clear that various embodiments in Williams teach that the node and its various qualities (materials, processes), are stored as vectorized word embeddings. In response to the applicant’s argument that Williams makes no mention of comparing word embeddings with another in a vector space, this argument is not persuasive because Williams directly makes references to comparing word embeddings in a vector space in [0159-0160]. Furthermore, even assuming arguendo this were the case, the claims are currently rejected under 35 U.S.C. 103 as obvious over Williams in view of Russo, and attacking any reference individually where the rejection is based on a combination of references is not proper.
As a result of the updated rejection, the applicant’s arguments following “Secondly,..” are also not persuasive because Williams alone is not relied upon to teach the storage of the embedding with a key that maps the at least one word embedding to the at least one emission factor. The combination of Williams and Russo teach or suggest this limitation, therefore this applicant’s argument is moot. Therefore, even when considering all of the applicant’s arguments, the claims remain rejected under prior art. Furthermore, the applicant’s arguments under C. Rejection under 35 U.S.C. 103 is also moot as the claims are now rejected under the combination of Williams and Russo instead. Therefore, claims 1-9 and 12-20 remain rejected under 35 U.S.C. 103.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NICO LAUREN PADUA whose telephone number is (703)756-1978. The examiner can normally be reached Mon to Fri: 8:30 to 5:00pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Jessica Lemieux can be reached at (571) 270-3445. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/NICO L PADUA/ Junior Patent Examiner, Art Unit 3626
/SANGEETA BAHL/ Primary Examiner, Art Unit 3626