DETAILED ACTION
Notices to Applicant
This communication is a final rejection. Claims 1-20, as filed 03/23/2026, are currently pending and have been considered below.
Priority is generally acknowledged to 63/508,441 which was filed 06/15/2023.
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon and the rationale supporting the rejection would be the same under either status.
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.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-19 are rejected under 35 U.S.C. 103 as being unpatentable over Leifer (US20190228856A1) in view of Mysore (“Large Language Model Augmented Narrative Driven Recommendations”).
Regarding claim 1, Leifer discloses: A dietary modification computing system (“parallel computing to improve system processing ability for improving food-related personalization across a plurality of users, user food-related preferences, SMEs, etc.)” [0013]) comprising:
--a taste-restrictions combination neural network configured for: receiving entity restriction profile data and entity taste profile data (“dietary preferences,” [0019]; “Determining user food-related preferences is preferably based on user inputs,” [0021]); and generating an entity taste-restrictions vector based on a combination of the entity restriction profile data and the entity taste profile data, the entity taste-restrictions vector including a first high-dimensional data structure in which a first set of data values describe taste-restrictions relationships among the entity restriction profile data and the entity taste profile data (“Vector representations of food-related data are preferably generated as a result of processing the food-related data at a trained synthetic neural network defining a plurality of neuronal layers,” [0018]; “the method can include deriving a recipe vector representation based on the user food preferences database, including: training a neural network model using the food preferences as inputs, wherein the neural network model is made up of a plurality of neuronal layers, and wherein the recipe vector representation is generated from (e.g., equivalent to) an intermediate layer (e.g., the weights associated with the intermediate layer) of the plurality of neuronal layers,” [0022]);
--a recipe-restrictions combination neural network configured for: receiving the entity restriction profile data and recipe data; and generating an entity recipe-restrictions vector based on a combination of the entity restriction profile data and the recipe data, (representing recipes as vectors and comparing them to user constraints like dietary restrictions [0016]-[0017]; );
--an entity-recipe relational recommendation module configured for: calculating a degree of multi-dimensional similarity among the entity taste-restrictions vector and the entity recipe-restrictions vector; and generating entity-recipe prediction data that describes the degree of (scores in [0012] and [0019]; “recipe matching scores indicating the degree to which user food-related preferences are satisfied by a given recipe,” [0012]; “Vector representations of food-related data (e.g., recipe data structures, ingredients or ingredient entities, user food preferences, etc.) can function to enable comparison of such data (e.g., with like data, with similar data, with subsets of partially similar data, etc.) across a vector of data features,” [0016]); and
--an entity-recipe relational modification module configured for: generating optimization data identifying, for a portion of the recipe data, an impact of the portion of the recipe data on the(modified meal option in [0030]-[0031]; [0039]; S140 in FIG. 1);
--wherein the dietary modification computing system is configured for: based on the optimization data, identifying substitution recipe data associated with the portion of the recipe data (“with recipe item substitutions applied to the base recipe,” [0031]);
--generating modified recipe data that includes a combination of the recipe data with the substitution recipe data, wherein the modified recipe data omits the portion of the recipe data identified by the optimization data; and providing the modified recipe data to a user computing device (“Personalized food parameters are preferably presented at a user interface,” [0037]).
Leifer does not expressly disclose, but Mysore teaches:
--the entity recipe-restrictions vector including a second high-dimensional data structure in which a second set of data values describe recipe-restrictions relationships among the entity restriction profile data and the recipe data (Bi-encoder models embed the query and item independently into high dimensional vectors on page 5);
--generating entity-recipe prediction data that describes the degree of multi-dimensional similarity (“rank items for the user based on the minimum L2 distance between qu and di” page 5);
It would have been obvious before the effective filing date to expand Leifer’s vector based food personalization techniques to include Mysore’s bi-encoder model techniques because these would “improve recommendation performance” (page 3) and allow for more accurate outputs in response to narrative queries (Fig. 1; page 2).
Regarding claim 2, Leifer further discloses: wherein: the entity restriction profile data includes restriction data values describing one or more dietary restrictions associated with an entity, and the entity taste profile data includes taste preference data values describing one or more taste preferences associated with the entity (determining user food-related preferences (e.g., health goals, taste preferences, dietary restrictions, etc.) associated with one or more users [0008]).
Regarding claim 3, Leifer does not expressly disclose but Mysore further teaches: wherein: the taste-restrictions combination neural network is prevented from accessing the recipe data, and the recipe-restrictions combination neural network is prevented from accessing the entity taste profile data (Bi-encoder models embed the query and item independently into high dimensional vectors on page 5; the Examiner further notes that ).
The motivation to combine is the same as in claim 1.
Regarding claim 4, Leifer further discloses: wherein the entity-recipe prediction data includes probability data indicating a relative effectiveness of the recipe data in addressing the combination of the entity restriction profile data and the entity taste profile data (“similarity scores between types of ingredient entities; recipe matching scores indicating the degree to which user food-related preferences are satisfied by a given recipe,” [0012]).
Regarding claim 5, Leifer further discloses: wherein the entity-recipe relational recommendation module generates the entity-recipe prediction data based on a principal component analysis (“PCA”) (principal component analysis in [0033]).
Regarding claim 6, Leifer further discloses: wherein the entity-recipe relational modification module generates the optimization data responsive to determining that the entity-recipe prediction data fulfills a particular relationship with an optimization threshold value (“Block S130 can include determining a base meal option (e.g., corresponding to a base recipe) for a user (e.g., satisfying a subset of the user food-related preferences; etc.); and determining a modified meal option (e.g., better suited for achieving user food-related goals; meeting other user food-related preferences; with recipe item substitutions applied to the base recipe, such as based on ranked food substitutability outputs from a food substitution model; satisfying all of the user food-related preferences; associated with a calculated higher probability of meeting user food-related preferences; etc.) based on the base meal option and dietary inputs (and/or other suitable data),” [0031]).
Regarding claim 7, Leifer further discloses wherein the substitution recipe data is received from an additional computing system (food modifications are come from a “food substitution model” and “food substitution database” [0014] and [0031]).
Leifer does not expressly disclose but Mysore further teaches: configured to provide one or more of a large language model (“LLM”) or an image generation model (LLM in the abstract and throughout).
One of ordinary skill in the art before the effective filing date would have found it obvious to implement Leifer’s food substitution model as an LLM as taught by Mysore because LLMs are able to provide accurate recommendations based on “strong language understanding capabilities” (page 2).
Claims 8-13 are analogous to claims 1-2 and 4-7 and are rejected with the same reasoning.
Claims 14-19 are analogous to claims 1-2 and 4-7 and are rejected with the same reasoning.
Regarding claim 20, Leifer further discloses: wherein the entity-recipe relational recommendation module calculates the degree of multi-dimensional similarity using principal component analysis (PCA) to reduce dimensionality of the entity taste-restrictions vector and the entity recipe-restrictions vector (principal component analysis for food-related vector representations in [0033]; “Vector representations of food-related data (e.g., recipe data structures, ingredients or ingredient entities, user food preferences, etc.) can function to enable comparison of such data (e.g., with like data, with similar data, with subsets of partially similar data, etc.) across a vector of data features,” [0016]).
Leifer does not expressly disclose that this calculation occurs prior to similarity calculation. Mysore teaches this. Mysore’s bi-encoder model “embed[s] the query and item independently into high dimensional vectors” on page 5. This means that the embeddings are formed before any similarity metric is computed and thus embedding compression precedes the distance calculation.
It would have been obvious before the effective filing date to expand Leifer’s vector based food personalization techniques to include applying PCA to reduce the vectors before computing similarity as in Mysore’s bi-encoder model because these would “improve recommendation performance” (page 3) and allow for more accurate outputs in response to narrative queries (Fig. 1; page 2).
Response to arguments
Applicant's arguments filed 03/23/2026 have been fully considered and are discussed below.
Regarding the prior art rejections, Applicant argues that the claimed dual neural network architecture is distinct from Leifer’s single unified neural network. “Leifer’s unified system processes both user preferences and recipe data through the same network architecture, which is fundamentally different from the claimed system that requires two separate neural networks with opposite data restrictions.” Remarks pages 9-10. This argument is not persuasive. The claims do not recite separately trained models, physically distinct hardware, or any other structural requirement that would preclude a single network architecture from reading on the claim language. The BRI of “a taste-restrictions combination neural network configured for receiving entity restriction profile data and entity taste profile data” is simply any neural processing component that takes restriction data and taste data as inputs. Leifer does this when it processes “dietary preferences” and “user food-related preferences” including “taste preferences, dietary restrictions, etc” in [0008] and [0019]. The Bri of “a recipe-restrictions combination neural network configured for: receiving the entity restriction profile data and recipe data” is likewise any neural processing component that takes restriction data and recipe data as inputs. Leifer does this when it “deriv[es] a recipe vector representation based on the user food preferences database” and compares vectors against dietary constraints in [0016], [0017], and [0022]. Under the BRI, a model invoked once with taste and restriction inputs and again with recipe and restriction inputs satisfies the two “combination neural networks” limitations of claim 1.
Regarding claim 3 where “the taste-restrictions combination neural network is prevented from accessing the recipe data, and the recipe-restrictions combination neural network is prevented from accessing the entity taste profile data,” applicant suggests a distinction from Leifer and Mysore deriving from some kind of gate or enforcement barrier. The BRI of the claim does not require this. “Prevented from accessing” includes any condition under which a network cannot access the data in question. The most straightforward way to implement this with a neural network is to not provide the data as in input. If recipe data is not in the input vector fed to the taste-restrictions network, that network is, under the BRI, prevented from accessing the data. No affirmative barrier is required. Additionally, Mysore’s bi-encoder models “embed the query and item independently into high dimensional vectors” on page 5 which is another form of “prevention” under the BRI of claim 3.
Applicant argues that Mysore is non-analogous art that cannot be combined with Leifer because the subject matter of Mysore is point-of-interest recommendations while Leifer’s is food recommendation. Remarks pages 10-11. This is not persuasive because the field of applicant’s endeavor is properly characterized as neural-network based information retrieval and recommender systems, not the narrower field of food recommendation. Both Leifer and Mysore use these techniques, and it is immaterial that Leifer makes recipe recommendations and Mysore makes point-of-interest recommendations. Applicant’s argument would, if accepted, limit the scope of prior art to only food or recipe recommendations and would render non-analogous recommender systems functioning on any other type of data. Additionally, Mysore is pertinent to the problem of independently encoding two different data types into separate high-dimensional vectors for similarity scoring. That problem is the same whether the data is points-of-interest or recipes. The Examiner notes that software is a highly predictive field of art. A bi-encoder that improves retrieval accuracy for point-of-interest queries would predictively improve retrieval accuracy for recipe queries.
Finally, applicant argues that there is no motivation to “split Leifer's working unified neural network system into two specialized networks with access restrictions, particularly when such modification would require substantial re-engineering without apparent benefit,” Remarks page 11. This argument is not persuasive. First, as described above in this section, the Examiner disagrees that Leifer’s must be “split” to read on the BRI of the claim language. Even if it were split, Mysore supplies the motivation, namely, that bi-encoder models “improve recommendation performance” on page 3 by allowing independent optimization of the query and item embedding spaces. Applicant’s assertion that Leifer acts on structured data and Mysore on unstructured data overstates the technical divergence between the systems and conflates the input data format with the architectural techniques. Mysore’s bi-encoder architecture is data-agnostic that is equally applicable whether the inputs are food preference vectors or narrative text.
Regarding claims 4 and 13, applicant argues that Leifer discloses only simple matching scores rather than the claimed probability data indicating relative effectiveness. Remarks page 12. This is not persuasive because the BRI of “probability data indicating a relative effectiveness” includes any score that quantifies how well a recipe satisfies a user’s preferences which Leifer discloses in [0012]: “recipe matching scores indicating the degree to which user food-related preferences are satisfied by a given recipe.”
Regarding claims 6, 12, and 18, applicant argues that neither reference teaches generating optimization data responsive to a threshold relationship with a prediction value, but this is not persuasive because Leifer discloses various techniques evaluating whether meal options “mee[t] user food-related preferences”.
Regarding claims 7, 13, and 19, applicant argues that the art does not teach image generation, but this is not persuasive because this limitation is optional as drafted. Yi’s Figure 2 shows that each tower receives only its designated input data, i.e., each tower is prevented from seeing the inputs from the other.
Conclusion
Prior art made of record but not relied upon for any rejections includes Yi (Xinyang Yi et al.; Sampling-bias-corrected neural modeling for large corpus item recommendations. In Proceedings of the 13th ACM Conference on Recommender Systems (RecSys '19). Association for Comput.). Yi is a recommender system with a feed-forward DNN that encodes user and context features such as watch history. Yi discloses “a modeling framework using two-tower neural net, with one of the towers (item tower) encoding a wide variety of item content features,” Abstract.
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 JOSHUA BLANCHETTE whose telephone number is (571)272-2299. The examiner can normally be reached on Monday - Thursday 7:30AM - 6:00PM, EST.
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, Shahid Merchant, can be reached on (571) 270-1360. 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.
/JOSHUA B BLANCHETTE/ Primary Examiner, Art Unit 3624
Prosecution Insights