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 .
Response to amendment
This office action is in response to an amendment filed on June 30, 2026 in response to PTO office action dated March 31, 2026.
Claims 1, 11 and 20 have been amended. Claims 1-20 are pending. The amendment has been entered and considered.
Applicant’s amendment to the Abstract with respect to the objection of the Abstract has been fully considered. As a result, the objection has been withdrawn.
Applicant’s amendment to the claim with respect to the rejection of the claims under 35 U.S.C. § 101 has been fully considered. As a result, the rejection has been withdrawn.
Applicant’s arguments with respect to the rejection of claims under 35 U.S.C. § 103(a) have been fully considered bur are moot in view of the new grounds of rejection.
This action is FINAL.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on June 30, 2026 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement has been considered by the examiner.
Claims rejection 35 U.S.C. 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 of this title, 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 1-7, 11-17 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Sandler et al. (US 11004135 B1) in view of Zhang et al. (CN 116341634 B) further in view of Beauchamp et al. (US 20240289365 A1).
Regarding claims 1, 11 and 20 Sandler discloses a computer-implemented method comprising:
receiving user input at a computing system comprising a processor and a memory (see Sandler col. 6, lines 55-67, The vector embedding generator 220 is an engine that takes user profile features or item data as inputs);
identifying . by the processor, contextual information associated with the user
converting, by the processor, the user input into a high-dimensional embedding (see Sandler col. 6, lines 55-67, The vector embedding generator 220 is an engine that takes user profile features or item data as inputs and converts these inputs into a high-dimensional vector representation of the input data, he vector embedding generator 220 can embed words into vectors from text using the Continuous Bag-of-Words model (CBOW), the Skip-Gram model, or other suitable word-vector embedding models. One example of these models implements a neural network architecture having an output neuron for each word in the vocabulary of the vector embedding generator 220, where the vocabulary can be provided to the vector embedding generator 220 or learned from a corpus of training text);
performing, by the processor, … using the high-dimensional embedding to retrieve and refine a search result of recommended items; and presenting, by the processor, the recommended items to a user (see Sandler col. 5, lines 1-10, a model as described herein may be trained periodically, such as once per day, and then used by a recommendation engine to generate recommendations for users as they interact with the electronic catalog. In some embodiments the model can be used offline to generate a user-specific list including items and associated probabilities of the user making a purchase of the item in an upcoming time interval. Such lists can be ranked based on the probability scores and subsequently provided to a diversity model as described herein in order to provide relevant and diverse recommendation sets to the user).
Zhang expressly discloses performing…a hierarchical search (see Zhang performing neural structure search model training is a process of performing neural network search, in which a hierarchical search similar to a previous work is used, and in particular, it is composed of three stages: a first stage, a second stage, a third stage, a third stage, a fourth stage, a fourth stage, and a fourth stage).
It would have been obvious to a person of ordinary skill in art before the effective filing date of the claimed invention to incorporate the teaching of Zhang into the method of Sandler to have performing…a hierarchical search. Here, combining Zhang with Sandler, which are both related to data processing, improves Sandler, by providing systems for performing optimization aiming at the reasoning time of the neural network, to solve the optimization for perfecting the reasoning time of the Neural architecture search (NAS) (see Zhang page 2).
Beauchamp converting using a customized transformer-based multi-modal embedding model that applies at least one of a convolutional neural network (see Beauchamp paragraph [0057], the encoder 52 and the decoder 54 each include a plurality of neural network layers, at least one of which may be a self-attention layer. The parameters of the neural network layers may be referred to as the parameters of the language model) or a vision transformer to extract visual features from image input (see Beauchamp paragraph [0045], DNNs are often used as ML-based models for modeling complex behaviors (e.g., human language, image recognition, object classification, etc.) in order to improve accuracy of outputs (e.g., more accurate predictions) such as, for example, as compared with models with fewer layers) and a natural language processing (NLP) transformer to extract semantic features from text input, and combines the extracted features using an embedding fusion technique into a shared space in which semantically or visually items are close together (see Beauchamp paragraph [0061], a short sequence of tokens 56 corresponding to the text sequence “Come here, look!” is illustrated as input to the transformer 50. Tokenization of the text sequence into the tokens 56 may be performed by some pre-processing tokenization module such as, for example, a byte pair encoding tokenizer (the “pre” referring to the tokenization occurring prior to the processing of the tokenized input by the LLM), which is not shown in FIG. 9 for simplicity. In general, the token sequence that is inputted to the transformer 50 may be of any length up to a maximum length defined based on the dimensions of the transformer 50 (e.g., such a limit may be 2048 tokens in some LLMs). Each token 56 in the token sequence is converted into an embedding vector 60 (also referred to simply as an embedding). An embedding 60 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 56; see Beauchamp paragraph [0092], An embeddings engine, such as the embeddings module 116 of FIG. 1, may be adapted to create the first vector embedding based on one or both of the search query and the input enhancement data. In at least some implementations, the search query input and the input enhancement data may be combined prior to creating the first vector embedding. That is, the first vector embedding may be created based on a combination of the search query and the input enhancement data. Alternatively, either the search query or the input enhancement data (but not both) may be embedded to create the first vector embedding).
It would have been obvious to a person of ordinary skill in art before the effective filing date of the claimed invention to incorporate the teaching of Beauchamp into the method of Sandler to have converting using a customized transformer-based multi-modal embedding model that applies at least one of a convolutional neural network. Here, combining Beauchamp with Sandler, which are both related to data processing, improves Sandler, by providing systems for performing Vector search that is useful for searching unstructured data, such as text, video, images, audio, etc., to identify content that is similar to a given query object (see Beauchamp paragraph [0003]).
Regarding claims 2 and 12 Sandler discloses wherein the user input includes one or more of an image, a text input, a categorical filter, or a voice input (see Sandler col. 21, lines 60-63, the interactive computing system 500 may also include input/output (I/O) device(s) and/or ports 514, such as for enabling connection with a keyboard, a mouse, a pen, a voice input device, a touch input device, a display, speakers, a printer, or other I/O device).
Regarding claims 3 and 13 Sandler discloses wherein performing the hierarchical search comprises:
performing a first search on a database to retrieve a set of content items based on measuring similarity between the high-dimensional embedding and embeddings of items stored in the database (see Zhang performing neural structure search model training is a process of performing neural network search, in which a hierarchical search similar to a previous work is used, and in particular, it is composed of three stages: a first stage, a second stage, a third stage, a third stage, a fourth stage, a fourth stage, and a fourth stage);
refining the set of content items by performing a second search on the set of content items based on at least one of a type of the user input or a number of the user input (see Zhang page 7, performing neural structure search model training is a process of performing neural network search, in which a hierarchical search similar to a previous work is used, and in particular, it is composed of three stages: a first stage, a second stage, a third stage, a third stage, a fourth stage, a fourth stage, and a fourth stage); and
performing a third search on the refined content items to determine the search result of recommended items using a parameterized algorithm operating in a continuous learning environment (see Zhang page 7, performing neural structure search model training is a process of performing neural network search, in which a hierarchical search similar to a previous work is used, and in particular, it is composed of three stages: a first stage, a second stage, a third stage, a third stage, a fourth stage, a fourth stage, and a fourth stage; see Gupta page 14, the artificial intelligence is a subject that makes the computer simulate some human thinking processes and intelligent behaviours (such as learning, reasoning, thinking, planning and so on), and both the hardware layer technology and the software layer technology. The artificial intelligence hardware technology generally comprises the technology such as sensor, special artificial intelligence chip, cloud computing, distributed storage, large data processing and so on; The artificial intelligence software technology mainly comprises computer vision technology, speech recognition technology, natural language processing technology and machine learning/deep learning, big data processing technology, knowledge map technology and so on).
It would have been obvious to a person of ordinary skill in art before the effective filing date of the claimed invention to incorporate the teaching of Zhang into the method of Sandler to have performing…a hierarchical search. Here, combining Zhang with Sandler, which are both related to data processing, improves Sandler, by providing systems for performing optimization aiming at the reasoning time of the neural network, to solve the optimization for perfecting the reasoning time of the Neural architecture search (NAS) (see Zhang page 2).
Regarding claims 4 and 14 Sandler discloses wherein …search comprise one or more of an image-to-image retrieval, an image-to-text retrieval, a text-to-image retrieval, or a text-to-text retrieval (see Sandler other keywords relating to item attributes can be used to generate vectors, with such attributes including the title and description as well as category, user reviews, and browse nodes of items in an electronic catalog, image features for image and video content, waveform features for audio and video content, lyrical features of music, and the like. In a 1-of-N (or “one-hot”) encoding, a vocabulary of relevant words is defined, and then every element in the vector is associated with a word in the vocabulary).
Zhang expressly discloses a hierarchical search (see Zhang performing neural structure search model training is a process of performing neural network search, in which a hierarchical search similar to a previous work is used, and in particular, it is composed of three stages: a first stage, a second stage, a third stage, a third stage, a fourth stage, a fourth stage, and a fourth stage).
It would have been obvious to a person of ordinary skill in art before the effective filing date of the claimed invention to incorporate the teaching of Zhang into the method of Sandler to have performing…a hierarchical search. Here, combining Zhang with Sandler, which are both related to data processing, improves Sandler, by providing systems for performing optimization aiming at the reasoning time of the neural network, to solve the optimization for perfecting the reasoning time of the Neural architecture search (NAS) (see Zhang page 2).
Regarding claims 5 and 15 Sandler discloses, wherein the one or more AI models include a customized transformer-based multi-modal embedding model (see Sandler col. 6, lines 23-35, the recommendation engine 200 includes a user profile features data repository 205, item data repository 260, diversified relevant recommendations data repository 230, vector embedding generator 220, relevance model 210, diversity model 225, and diversified relevant recommendations data repository 230. One or both of the relevance model 210 and diversity model 225 can be a machine learning model, and the recommendation engine 200 can be considered as implementing a model ensemble to provide recommendations).
Regarding claims 6 and 16 Sandler discloses comprising: identifying contextual information associated with the user input; and appending the identified contextual information to the user input, wherein both the user input and the appended information are converted into the high dimensional embedding (see Sandler col. 2, lines 51-60, relevance model includes a feedforward neural network architecture with an input feature set that includes a variety of features in order to make the network robust to the “cold start” problem (e.g., where a user has little or no associated behavioral data/purchase data). In the context of the pantry catalog, there can be a range of thousands of different input features including pantry catalog items purchased by a user, consumable items from other areas of the electronic catalog outside the pantry catalog that were purchased by the user, the brands of past purchases, item-type keywords of the past purchases, departments of products previously bought by the user, and high-level product categories such as general ledger categories of the products previously bought by the user).
Regarding claims 7 and 17 Sandler discloses wherein the search result of recommended items includes a first item and a second item, and the first item is presented to the user before the second item, and the method further comprising:
dynamically updating the second item based on user interaction with the first item (see Sandler col. 5, lines 35-45, The disclosed machine learning system can address this problem by presenting personalized, relevant, and diverse recommendations of the qualifying items. As will be appreciated, the disclosed machine learning system architecture can be extended to recommending relevant-diverse sets of other pools of qualifying items, for example promotional digital media content, recommendations of users or user groups on a social media site; see Sandler col. 16, lines 51-60, The main update equation in the GDPP (block 4, solving for j*) is the same as equation (4). Rather than summing objective scores across all possible sets of cardinality k, the update equation is performed at each stage to consider the existing recommendation set together with each of the remaining items in the candidate pool to identify the next item to add to the set).
Claims 8-9, and 18-19 are rejected under 35 U.S.C. 103 as being unpatentable over Sandler et al. (US 11004135 B1) in view of Zhang et al. (CN 116341634 B) in view of Beauchamp et al. (US 20240289365 A1) further in view of Haq et al. (US 12105729 B1).
Regarding claims 8 and 18 Haq expressly discloses determining one or more user permissions and roles for the user; and verifying an identity of the user based on the user permissions and roles (see Haq col. 13, lines 62-col. 14, lines 1-5; the system securely stores the user's registration information, which can include, but are not limited to, an assigned role (e.g., user, guest, manager, administrator), permissions (e.g., read, edit, create, write, delete), and access rights (e.g., personal domain, group domain, all domains), in the governed search system 200 database. Alternatively, existing users can sign in 501 and undergo a user authentication process, where they must enter their credentials (for example, username and password) to verify their identity).
It would have been obvious to a person of ordinary skill in art before the effective filing date of the claimed invention to incorporate the teaching of Haq into the method of Sandler to have determining one or more user permissions and roles for the user. Here, combining Haq with Sandler, which are both related to data processing, improves Sandler, by providing systems for authentication of users where they must enter their credentials (for example, username and password) to verify their identity (see Haq col. 14, lines 1-5).
Regarding claims 9 and 19 Sandler discloses providing a customized landing page to the user based …through a graphical user interface; and adding one or more interactive elements in the graphical user interface to receive the user input from the user (see Sandler As shown by the bolded “Pantry” illustrated in the selectable catalog sections 420, the user is visiting the landing page of the pantry catalog section of a larger electronic catalog. As such, the recommendations carousel 415 presents representations of five pantry catalog items, identified as both relevant to the user and diverse to one another as described herein. Each recommendation includes an image of the item, the name of the item, a star rating of the item, the price of the item, and a selectable feature to add the item to the user's digital shopping cart. The carousel 415 includes navigation controls 410A, 410B for scrolling through additional recommendations. As such, only a portion of the determined set of relevant, diversified items may be presented to the user at a time, with the most relevant item presented first).
Haq expressly discloses user permissions (see Haq col. 13, lines 62-col. 14, lines 1-5; the system securely stores the user's registration information, which can include, but are not limited to, an assigned role (e.g., user, guest, manager, administrator), permissions (e.g., read, edit, create, write, delete), and access rights (e.g., personal domain, group domain, all domains), in the governed search system 200 database. Alternatively, existing users can sign in 501 and undergo a user authentication process, where they must enter their credentials (for example, username and password) to verify their identity).
It would have been obvious to a person of ordinary skill in art before the effective filing date of the claimed invention to incorporate the teaching of Haq into the method of Sandler to have determining one or more user permissions and roles for the user. Here, combining Haq with Sandler, which are both related to data processing, improves Sandler, by providing systems for authentication of users where they must enter their credentials (for example, username and password) to verify their identity (see Haq col. 14, lines 1-5).
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Sandler et al. (US 11004135 B1) in view of Zhang et al. (CN 116341634 B) in view of Beauchamp et al. (US 20240289365 A1) further in view of Gupta et al. (US 11861688 B1).
Regarding claim 10, Gupta expressly discloses creating a plurality of shuffles to provide dynamic content (see Gupta col. 17, lines 20-30, the training data is shuffled before training, or between passes of the training. The shuffling in many embodiments is a random or pseudo-random shuffling to generate a truly random ordering, although there may be some constraints in place to ensure that there is no grouping of certain types of data, or the shuffled data may be reshuffled if such grouping exists, etc. Shuffling changes the order or arrangement in which the data is utilized for training so that the training algorithm does not encounter groupings of similar types of data, or a single type of data for too many observations in succession. For example, a model might be trained to predict a product type, where the training data includes movie, toy, and video game product types. The data might be sorted by product type before uploading. The algorithm can then process the data alphabetically by product type, seeing only data for a type such as movies first. The model will begin to learn patterns for movies).
It would have been obvious to a person of ordinary skill in art before the effective filing date of the claimed invention to incorporate the teaching of Gupta into the method of Sandler to have creating a plurality of shuffles to provide dynamic content. Here, combining Gupta with Sandler, which are both related to data processing, improves Sandler, by providing system in which the data is utilized for training so that the training algorithm does not encounter groupings of similar types of data, or a single type of data for too many observations in succession (see Gupta col. 17, lines 20-30).
Remarks
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Jiang et al. (US 11562328 B1) discloses the machine learning model can be any conventional embedding model for generating media embeddings (e.g., image embeddings). In some embodiments, the machine learning model includes a combination of at least a convolutional neural network layer and a plurality of multilayer perceptron layers. The convolutional neural network layer can apply one or more convolutions to inputted media (e.g., an image) to determine a corresponding feature vector that includes information describing the media. For example, the feature vector can include information for classifying the media and/or one or more objects represented in the media. The feature vectors outputted by the convolutional neural network layer can be provided to the plurality of multilayer perceptron layers for concatenation. The concatenated embeddings together represent a media embedding that corresponds to a second portion of the user embedding for the user.
Gong (US 20220164853 A1) discloses that after receiving an image query uploaded from a user, the uploaded image 250 is pre-processed (e.g., via channel-wise normalization, image resizing) and passed through one or more convolutional neural networks such as model A 252 and/or model B 254 to yield a high-dimensional image embedding 256 and 258. Both model A 252 and model B 254 (referred to as 352 and 354, respectively in FIG. 3) may be pre-trained on training sets of food-based images and combined to output an accurate classification of the received input. It is understood that there may be additional models as appropriate to obtain the final embedding. Example training sets include ImageNet, Recipe1m, Food101, and Menu-Match. Example neural networks include ResNet50, ResNeXt50, and DenseNet, and VGG, which may serve as model B 254. One such example neural network included approximately 25 million parameters and is convolutional neural network that is 50 layers deep.
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 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 date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DINKU W GEBRESENBET whose telephone number is (571)270-1636. The examiner can normally be reached between 8:00AM-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, Amy Ng can be reached on 571- 270-1698. 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.
/DINKU W GEBRESENBET/Primary Examiner, Art Unit 2164
Prosecution Insights