DETAILED ACTION
This Final Office Action is in response Applicant communication filed on
3/17/2026. In Applicant’s amendment, no claims were amended.
Claims 1-10 are currently pending and have been rejected as follows.
Response to Amendments
Rejections under 35 USC 101 are maintained.
Response to Arguments
Applicant’s 35 USC 101 rebuttal arguments and amendments have been fully considered but they are not persuasive to overcome the rejection.
Applicant argues on p. 5 that the withdrawal of the prior art rejections bears directly on the Step 2B analysis. Examiner respectfully disagrees. Lack of novelty under 35 U.S.C. 102 or obviousness under 35 U.S.C. 103 of a claimed invention does not necessarily indicate that additional elements are well-understood, routine, conventional elements. Because they are separate and distinct requirements from eligibility, patentability of the claimed invention under 35 U.S.C. 102 and 103 with respect to the prior art is neither required for, nor a guarantee of, patent eligibility under 35 U.S.C. 101. See MPEP 2106.05.I.
Applicant argues on p. 5 that “With respect to Step 2A, Prong One, the Office Action does not actually answer Applicant's prior argument. Applicant previously explained that the claimed operations, including embedding generation, positional encoding, multi-attention processing, and diversity optimization based on semantic distances and attention-derived relationships, are not mental steps and are not the sort of activity that can be performed in the human mind. The present Office Action does not substantively dispute that point. Instead, it shifts the alleged abstract idea from a mental-process theory to a "mathematical concepts" theory.” Examiner respectfully submits none of the previous office actions assert the claims are directed to the abstract idea grouping of mental processes. The response to arguments for the office action dated 5/13/2025 explained claims 1 and 5 similarly recite comparing thought objects to identify dissimilar thought objects, and selecting a thought object that maximizes semantic diversity, which are directed to mental processes because comparing and categorizing data is a concept that can be done in the human mind. That characterization is limited to those limitations. Further, the mere exclusion of one abstract idea grouping does not automatically make the claims patent eligible at Step 2A, Prong One when there is still another abstract idea recited.
Applicant argues on p. 5-7 that the present rejection does not analyze the claims as a whole and isolates individual features to label them as mathematical. Examiner respectfully disagrees. The limitations reciting mathematical relationships and calculations are identified in the Step 2A, Prong One analysis below. Identifying those limitations separately at Prong One is not improper dissection. The limitations are then considered together with the remaining elements at Step 2A, Prong Two. Applicant’s reliance on Desjardins at Step 2A, Prong One is misplaced because Desjardin itself found that the claim recited a mathematical concept at Prong One. Eligibility was found at Step 2A, Prong Two because the claim as a whole recited an improvement to machine learning itself, as described in the specification.
Applicant argues on p. 7 that “The Step 2A, Prong Two analysis is also incomplete because it does not meaningfully engage the specific technological improvement recited in the claims. Applicant's position is not that machine learning is patent-eligible merely because it occurs in a computer environment. Applicant's position is that the claims are directed to a specific improvement in the way the machine-learning system computes and applies semantic diversity. As Desjardins explains, claims are not ineligible merely because they are implemented in software or involve mathematical operations. Software-driven logical structures and processes can improve computer technology, and the question is whether the claim is directed to such an improvement rather than to the abstract idea itself. Here, the claims are directed to an improved way for a machine-learning-based selection system to generate, process, compare, and optimize embeddings for diversity-aware selection. That is a practical application.”Examiner respectfully disagrees. Applicant’s specification frames the problem the invention is solving as the difficulty deciding the next best qualitative response to display, providing sufficiently diverse content, obtaining equal coverage, and reducing bias ([0003]-0007]). The identified problem is not technical or computer related. The claimed solution does not rely on any specific technical architecture and that is demonstrated by the specification presenting alternative architectures for deploying the solution, Applicant’s Specification at least [0129]-[0134] discussing embeddings; [0150] discussing examples of decoder-based autoregressive models; [0152] discussing examples of base LLMs; and [0200]-[0202] discussing the range of possible modifications to the above. Further, the present claimed invention is not analogous to Desjardins. Desjardins addressed a technical problem of catastrophic memory loss when training the model on new tasks. The present claims apply an LLM to improve the usefulness of the content selection. Improving the usefulness of an output is an improvement to the abstract idea, not a technical improvement. The claims are also not similar to Enfish. Enfish’s self-referential table is a particular computer data structure that functioned differently than conventional database structures at the time, such as relational data structures and object oriented data structures. The self-referential table in Enfish improved how a computer stores and retrieves data in memory. In contrast, the present claims do not recite an improvement to computer or LLM functionality. The claimed embedding, positional encoding, and attention layers perform their ordinary NLP functions, while the calculation recited improve the information selection of diverse thought objects. The claims do not alter the technical architecture of any other technological function. Accordingly, the computer and LLM are invoked as tools for performing the mathematical analysis.
Applicant argues on p. 7-8 that the Step 2B analysis is deficient and improperly cites the withdrawal of the prior art rejection as evidence of unconventionality. As explained above, Lack of novelty under 35 U.S.C. 102 or obviousness under 35 U.S.C. 103 of a claimed invention does not necessarily indicate that additional elements are well-understood, routine, conventional elements. Because they are separate and distinct requirements from eligibility, patentability of the claimed invention under 35 U.S.C. 102 and 103 with respect to the prior art is neither required for, nor a guarantee of, patent eligibility under 35 U.S.C. 101. See MPEP 2106.05.I.Further, Applicant’s specification illustrates the conventionality of the additional elements employed to perform the abstract idea in [0045]-[0060] describing general purpose processors, memories, servers, networks, operating systems, interfaces, and databases; [0130]-[0160] describing standard embeddings, positional encoding, transformer attention, SoftMax, GPT, BERT; and [0185]-[0202] describing known embedding, aggregation, and semantic distance techniques. Further, applicant improperly conflates the limitations directed to the abstract idea with the additional elements. The additional elements’ interactions with each other as an ordered combination obtains inputs and executes the mathematical analysis (abstract idea) to return a result. The combination of the additional elements does nothing more beyond implementing the mathematical semantic diversity analysis.
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 1-10 are clearly drawn to at least one of the four categories of patent eligible subject matter recited in 35 U.S.C. 101 (systems and method). Claims 1-10 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without integrating the abstract idea into a practical application or amounting to significantly more than the abstract idea.
Regarding Step 1 of the 2019 Revised Patent Subject Matter Eligibility Guidance (‘2019 PEG”), Claims 1-4 and 9-10 are directed toward the statutory category of a machine (reciting a “system”). Claims 5-8 are directed toward the statutory category of a process (reciting a “method”).
Regarding Step 2A, prong 1 of the 2019 PEG, Claims 1 and 5 are directed to an abstract idea by reciting receive a plurality of thought objects, wherein the plurality of thought objects comprises text present in qualitative responses received from … wherein the plurality of thought objects comprises M thought objects that are most recently seen thought objects and N thought objects that are least seen … apply positional encoding to the embeddings to provide position information for each token in the thought objects … wherein the multiple attention layers compute token-wise contextual relationships that are used to adjust a diversity optimization metric derived from semantic distance calculations between embeddings: combine the processed token embeddings to generate a single embedding representation for each of the M thought objects and N thought objects; […] compare, the M thought objects and the N thought objects to identify one or more dissimilar thought objects in the N thought objects, wherein the comparison is performed based on semantic distances between the processed embeddings, wherein the semantic distances are calculated using at least one of cosine distance and Euclidean distance; and select one or more thought objects from the dissimilar thought objects as diverse thought objects by iteratively adjusting the selection based on a diversity optimization metric that maximizes orthogonality among embeddings in a high-dimensional semantic space. wherein maximizing semantic diversity comprises reducing redundancy among embeddings through a distance-based loss function computed over the attention-weighted relationships (Example Claim 1).
Claim 9 is directed to an abstract idea by reciting … receive a plurality of thought objects, wherein the plurality of thought objects comprises text present in qualitative responses received … generate word embedding for each thought object … and combine the generated word embeddings within a Siamese neural network that aggregates the embeddings using weighted averaging or attention pooling based on intra-object semantic relevance to produce an embedding per thought object; determine, based on semantic distances calculated using at least one of cosine distance and Euclidean distance between the embeddings, diverse thoughts present among the plurality of thought objects: select a subset of thought objects from the plurality of thought objects based on the determined diverse thoughts, wherein the selected subset maximizes semantic diversity by minimizing pairwise similarity among embeddings through a distance-based optimization metrics.
The claims are considered abstract because these steps recite mathematical concepts including mathematical relationships and mathematical calculations. The claims recite receiving thought objects, comparing the thought objects to identify different objects, and select a different object (claims 1 and 5). Claim 9 recites receiving thought objects, generating, and combining word embeddings. Claims 1, 5, and 9 recite steps diverse thought object selection using LLM and embedding models, which falls under mathematical concepts.
Regarding Step 2A, prong 2 of the 2019 PEG, the judicial exception is not integrated into a practical application because the claims (the judicial exception and the additional elements such as a thought selection server, comprising a processor, a memory, and a plurality of programming instructions, the plurality of programming instructions when executed by the processor cause the processor to; a plurality of user devices; process each of the M thought objects and N thought objects through an input processor and embedding layer of a Large Language Model (LLM) to generate embeddings, wherein the embedding layer converts discrete tokens into continuous vector representations; process the position-encoded embeddings through multiple attention layers of the LLM to identify relationships between tokens in the thought objects; provide, a prompt, to the LLM with the thought selection server, wherein the prompt comprises a request to identify diverse thought objects and comprises contextual metadata describing characteristics of the M and N thought objects whereby the LLM modifies subsequent diversity selections based on responses to the prompt; using a non-transformer based embedding model, comprising at least one of GloVe, Word2Vec, and FastText) are not an improvement to a computer or a technology, the claims do not apply the judicial exception with a particular machine, the claims do not effect a transformation or reduction of a particular article to a different state or thing nor do the claims apply the judicial exception in some other meaningful way beyond generally linking the use of the judicial exception to a particular technological environment such that the claims as a whole is more than a drafting effort designed to monopolize the exception (see MPEP §§ 2106.05(a-c, e)).
Dependent claims 2-4, 6-8, and 10 do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the limitations recite mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea ‐ see MPEP 2106.05(f).
Regarding Step 2B of the 2019 PEG, the additional elements have been considered above in Step 2A Prong 2. The claim limitations do not amount to significantly more than the judicial exception because they are directed to limitations referenced in MPEP 2106.05I.A. that are not enough to qualify as significantly more when recited in a claim with an abstract idea because the limitations recite mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea ‐ see MPEP 2106.05(f).
Applicant's claims mimic conventional, routine, and generic computing by their similarity to other concepts already deemed routine, generic, and conventional [Berkheimer Memorandum, Page 4, item 2] by the following [MPEP § 2106.05(d) Part (II)]. The claims recite steps like: “Receiving or transmitting data over a network, e.g., using the Internet to gather data,” Symantec, “Performing repetitive calculations,” Flook, and “storing and retrieving information in memory,” Versata Dev. Group, Inc. v. SAP Am., Inc. (citations omitted), by performing steps to “receive” and “process” thought objects, “apply” positional encodings, “process” the position-encoded embeddings, “combine” the processed token embeddings, “provide” a prompt, “compare” thought objects, and “select” one or more thought objects (Example Claim 1) and “generate” and “combine” word embeddings, “determine” diverse thoughts, and “select” a subset of thought objects (Example Claim 9).
The Berkheimer Memorandum, Page 4, item 3, states an additional element (or combination of elements) is well-understood, routine, or conventional if a citation to a publication demonstrates the well-understood, routine, conventional nature of the additional element(s). Previously cited Vaswani et al., Attention Is All You Need, 2017, hereinafter Vaswani does so for the additional elements. See Vaswani p. 5, section 3.4 “Embeddings and Softmax;” section 3.5 “Positional Encoding;” section 3.2.2 “Multi-Head Attention.” An improvement over produced (similarity) metrics found in prior art is not a test for patent eligibility under 35 USC 101. Further, employing a Siamese neural network “that aggregates embeddings through weighted or attention pooling, yielding a system configuration that is structurally distinct from generic data-analysis algorithms” is well-understood, routine, or conventional. See Neculoiu et al., Learning Text Similarity with Siamese Reccurent Networks, 2016, hereinafter Neculoiu, p. 149 where it explains the Siamese network architecture has been around since 1993.
By the above, the claimed computing “call[s] for performance of the claimed information collection, analysis, and display functions ‘on a set of generic computer components' and display devices” [Elec. Power Group, 830 F.3d at 1355] operating in a “normal, expected manner” [DDR Holdings, LLC v. Hotels.com, L.P., 773 F.3d at 1245, 1258 (Fed. Cir. 2014)].
Conclusively, Applicant's invention is patent-ineligible. When viewed both individually and as a whole, Claims 1-10 are directed toward an abstract idea without integration into a practical application and lacking an inventive concept.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
US 2017/0161279 A1: A method and apparatus are provided for recommending concepts from a first concept set in response to user selection of a first concept Ci by performing a natural language processing (NLP) analysis comparison of vector representations of user concepts contained in written content authored by the user and candidate concepts in a first concept set to determine a similarity measure for each candidate concept, and to select therefrom one or more of the candidate concepts for display as recommended concepts which are related to the user concepts contained in written content authored by the user based on the similarity measure between each candidate concept and each user concept.
WO 2016/102153 A1: The invention discloses a method implemented by computer for semantically describing the content of an image comprising the steps consisting of receiving a signature associated with said image; receiving a plurality of groups of initial visual concepts; the method being characterised by the steps consisting of expressing the signature of the image in the form of a vector comprising components referring to the groups of initial visual concepts; and modifying said signature by applying a filtering rule that is applied to the components of said vector. Developments describe, in particular, filtering rules that involve filtering by thresholds and/or by intra-group or inter-group order statistics, partitioning techniques including the visual similarity of the images and/or the semantic similarity of the concepts, and the optional addition of manual annotations to the semantic description of the image. The advantages of the method in terms of sparse and diversified semantic representation are presented.
Chen et al., A History and Theory of Textual Event Detection and Recognition, 2020: There is large and growing amounts of textual data that contains information about human activities. Mining interesting knowledge from this textual data is a challenging task because it consists of unstructured or semistructured text that are written in natural language. In the field of artificial intelligence, event-oriented techniques are helpful in addressing this problem, where information retrieval (IR), information extraction (IE) and graph methods (GMs) are three of the most important paradigms in supporting event-oriented processing. In recent years, due to information explosions, textual event detection and recognition have received extensive research attention and achieved great success. Many surveys have been conducted to retrospectively assess the development of event detection. However, until now, all of these surveys have focused on only a single aspect of IR, IE or GMs. There is no research that provides a complete introduction or a comparison of IR, IE, and GMs. In this article, a survey about these techniques is provided from a broader perspective, and a convenient and comprehensive comparison of these techniques is given. The hallmark of this article is that it is the first survey that combines IR, IE and GMs in a single frame and will therefore benefit researchers by acting as a reference in this field.
THIS ACTION IS MADE FINAL. 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 extension fee 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 MOHAMED EL-BATHY whose telephone number is (571)270-5847. The examiner can normally be reached on M-F 8AM-4:30PM.
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, PATRICIA MUNSON can be reached on (571) 270-5396. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/MOHAMED N EL-BATHY/Primary Examiner, Art Unit 3624
Prosecution Insights