Method, device and medium for providing personalized nutritional product recommendation
By obtaining the multi-dimensional data and behavioral data of users, generating user portraits, combining nutrition product knowledge graphs, using recall and sorting models, the inaccuracy problem of nutrition product recommendations in the existing technology is solved, and personalized nutrition product Accurate recommendation.
Patent Information
- Application Number
- CN202311492509.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-05-26
- Filing Date
- 2023-11-09
- Publication Date
- 2024-11-26
AI Technical Summary
The prior art is difficult to accurately judge health status based on user's multi-faceted data and provide personalized nutritional product recommendations, resulting in a lack of accuracy and effectiveness when taking nutritional products.
By obtaining the user's multi-dimensional physiological data and behavioral data, static and dynamic tags are generated, user portraits are built, and combined with the knowledge graph of nutritional products, using recall models and sorting models to generate personalized nutritional product recommendations.
It provides accurate and personalized nutritional product recommendations based on the user's health status and preferences, and improves the accuracy and effectiveness of nutritional product use.
Smart Images

Figure CN119028529A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to a computer-implemented method, device and medium for providing personalized nutritional product recommendations, and more specifically, to a computer-implemented method for providing personalized nutritional product recommendations based on user portraits and a knowledge graph of nutritional products. Background Art
[0002] At present, with the improvement of living standards, people are more and more concerned about their physical health. For example, people often have regular physical examinations to learn about their health status. In addition, with the emergence of wearable devices, people can also use wearable devices to monitor their exercise, sleep and other conditions. When the body has sub-health symptoms such as poor sleep, dull skin, and high stress, people may go to the doctor and take some nutritional products according to the doctor's advice to adjust their physical condition.
[0003] However, in many cases, although users are in sub-health conditions, they feel that it is troublesome to go to the hospital or the condition is not serious enough to require a visit, so they take some nutritional products based on their own experience. The type, quantity, and frequency of nutritional products are all quite arbitrary. This way of taking nutritional products is very inaccurate and often fails to alleviate the user's symptoms. This is because users judge their physical condition based on vague subjective feelings and cannot accurately determine what nutritional needs they have. In addition, different nutritional products contain different nutritional elements and target different symptoms. Ordinary users do not have the corresponding nutritional knowledge, so users cannot accurately determine which nutritional products to take based on their own status data.
[0004] In recent years, with the development of genetic testing technology, people can now directly learn what nutritional needs they have through genetic testing. However, there is no direct correlation between the nutritional needs obtained through such tests and physical conditions.
[0005] Therefore, how to accurately judge the user's health status based on the user's physical condition, mental state, genetic test and other data, and provide the user with personalized nutritional product recommendations conveniently and accurately has become an urgent problem that needs to be solved. Summary of the invention
[0006] Based on this, various embodiments of the present disclosure provide a method and apparatus for personalized nutritional product recommendation, so as to conveniently and accurately provide personalized nutritional product recommendations to users.
[0007] According to one aspect of the present disclosure, there is provided a computer-implemented method for providing personalized nutritional product recommendations, comprising:
[0008] Obtain multiple dimensions of physiological data information of users;
[0009] Generating a plurality of static tags indicating the health status of the user according to the physiological data information of the plurality of dimensions;
[0010] Obtain multiple behavioral data of users;
[0011] generating a plurality of dynamic tags indicating preferences of the user according to the plurality of behavior data;
[0012] Generate a user portrait based on the static tags and the dynamic tags;
[0013] Based on the multiple dynamic tags and the knowledge graph based on the nutritional products, multiple recall lists indicating the correspondence between users and different nutritional products are generated through the trained recall model, and a recall fusion list that fuses the multiple recall lists is constructed;
[0014] constructing a feature vector for a ranking model using the recall fusion list and the user portrait, and inputting the calculated feature vector into the ranking model; and
[0015] The recall fusion list is ranked according to the scores calculated by the ranking model on the input feature vectors to determine the recommendation of personalized nutritional products suitable for the user.
[0016] Preferably, the physiological data information of multiple dimensions includes one or more of the user's physical examination report data, the user's physiological data received from a wearable device, the user's self-reported nutritional needs data, and the user's genetic test report data.
[0017] Preferably, generating a plurality of static tags indicating the physiological state of the user according to the physiological data information of the plurality of dimensions comprises:
[0018] generating a first type of label with a first weight according to the user's physical examination report data, wherein the first type of label includes a plurality of labels indicating different physiological health states of the user;
[0019] generating a second type of tag with a second weight according to the physiological data of the user received from the wearable device, wherein the second type of tag includes a plurality of tags indicating different physiological activity states of the user;
[0020] generating a third type of tags with a third weight according to the nutritional requirement data reported by the user himself, wherein the third type of tags includes a plurality of tags indicating different mental states and nutritional requirements reported by the user himself;
[0021] generating a fourth type of label with a fourth weight according to the user's gene test report data, the fourth type of label including a plurality of labels indicating different long-term nutritional needs of the user; and
[0022] Generate multiple static tags indicating the health status of the user according to the user basic information, the first type of tags, the second type of tags, the third type of tags and the fourth type of tags,
[0023] The first weight, the second weight, the third weight and the fourth weight have decreasing values in sequence.
[0024] Preferably, the multiple behavior data of the user include user browsing records, user search records, user question and answer records, and corresponding time records.
[0025] Preferably, generating a plurality of dynamic tags indicating user preferences according to the plurality of behavior data comprises:
[0026] Generate browsing tags, search tags and question-and-answer tags based on user browsing records, user search records and user question-and-answer records of multiple users, wherein the tags include three types: entity, keyword and classification;
[0027] Building multiple collections of browsing tags, search tags, and question-and-answer tags of multiple users within a predetermined time period based on the user as a unit;
[0028] Calculating a score for each user tag in the plurality of sets using a term frequency-inverse document frequency (TF-IDF) algorithm;
[0029] Generate the final score of each user tag based on the score of each user tag, the time decay of the interaction, the weight of the interaction type, and the weight of the tag type;
[0030] A predetermined number of user tags with the highest final scores are selected as dynamic tags of the users.
[0031] Preferably, the recall model includes:
[0032] An SVD recall sub-model based on a singular value decomposition (SVD) model, for generating an SVD recall list indicating the products that the user is most likely to purchase;
[0033] an update recall sub-model for generating an update recall list indicating the most recently updated products;
[0034] The hot recall sub-model is used to generate a hot recall list of popular products based on the product's click volume and time decay;
[0035] A graph recall sub-model is used to query products associated with the tags from the nutritional product-based knowledge graph according to multiple tags of the user, and generate a graph recall list of personalized products; and
[0036] The fusion sub-model is used to fuse the SVD recall list, the update recall list, the hot recall list and the atlas recall list according to a predetermined fusion rule to generate a recall fusion list.
[0037] Preferably, constructing a feature vector for a ranking model using the recall fusion list and the user portrait includes:
[0038] Performing feature engineering on each information field in the user portrait; and
[0039] According to the weights of the various features, a weight assignment process is performed to generate a feature vector for a sorting model, wherein the weights in the static labels are greater than the weights in the dynamic labels.
[0040] Preferably, sorting the plurality of recall lists according to the scores calculated by the sorting model on the input feature vectors to determine the recommendation of personalized nutritional products suitable for the user comprises:
[0041] The ranking model calculates a score indicating the user's interest in the product based on the input feature vector;
[0042] Sort the multiple recall lists contained in the recall fusion list according to the value of the score; and
[0043] The nutritional product in the recall list with the highest score is determined as a recommendation of a personalized nutritional product suitable for the user.
[0044] Preferably, the personalized nutritional product includes a combination product unit of a plurality of different types and different quantities of nutritional products corresponding to the health status of the user.
[0045] Preferably, the method further comprises:
[0046] Determine the user's health status based on the user's static tags;
[0047] Obtaining related nutritional products based on the user's health status and the knowledge graph of nutritional products;
[0048] Based on the obtained associated nutritional products, a recommendation of related nutritional products suitable for the user is determined.
[0049] Preferably, the method further comprises:
[0050] Displaying the personalized nutritional product recommendation and / or the related nutritional product recommendation to the user; or
[0051] sending the personalized nutritional product recommendation and / or the related nutritional product recommendation to another electronic device, and
[0052] The recommendation of the personalized nutritional product and / or the recommendation of the related nutritional product are displayed on another electronic device.
[0053] Preferably, the method further comprises:
[0054] Determine the user's health status based on the user's static tags;
[0055] The determined health status and health advice corresponding to the health status are displayed to the user.
[0056] Preferably, the method further comprises:
[0057] Determine the user's nutritional needs based on the user's static tags;
[0058] Obtaining related nutritional products based on the user's nutritional needs and the knowledge graph of nutritional products;
[0059] Based on the obtained associated nutritional products, a recommendation of related nutritional products suitable for the user is determined.
[0060] Preferably, the method further comprises:
[0061] Displaying to the user the nutritional needs of the user and / or recommendations of the personalized nutritional products; or
[0062] sending the personalized nutritional product recommendation and / or the related nutritional product recommendation to another electronic device, and
[0063] The recommendation of the personalized nutritional product and / or the recommendation of the related nutritional product are displayed on another electronic device.
[0064] According to another aspect of the present invention, there is provided an apparatus for providing personalized nutritional product recommendations, comprising:
[0065] a memory having a computer program stored thereon; and
[0066] The processor is configured to, when executing the computer program, perform:
[0067] Obtain multiple dimensions of physiological data information of users;
[0068] generating a plurality of static tags indicating the health status of the user according to the physiological data information of the plurality of dimensions;
[0069] Obtain multiple behavioral data of users;
[0070] generating a plurality of dynamic tags indicating preferences of the user according to the plurality of behavior data;
[0071] Generate a user portrait based on the static tags and the dynamic tags;
[0072] Based on the multiple dynamic tags and the knowledge graph based on the nutritional products, multiple recall lists indicating the correspondence between users and different nutritional products are generated through the trained recall model, and a recall fusion list that fuses the multiple recall lists is constructed;
[0073] constructing a feature vector for a ranking model using the recall fusion list and the user portrait, and inputting the calculated feature vector into the ranking model; and
[0074] The recall fusion list is ranked according to the scores calculated by the ranking model on the input feature vectors to determine the recommendation of personalized nutritional products suitable for the user.
[0075] According to another embodiment of the present invention, a non-volatile storage medium is provided, on which a computer program is stored. When the program is executed by a computer, the method in the previous embodiment is executed.
[0076] According to the method and apparatus for providing personalized nutritional product recommendations disclosed herein, it is possible to analyze the user's data with the help of data modeling, machine learning, and medical knowledge, thereby providing personalized insights and recommendations for the user. In addition, according to the method and apparatus for providing personalized nutritional product recommendations disclosed herein, it is also possible to obtain recommended nutritional requirements from genetic tests, rather than directly investigating the original genetic test data. In addition, nutritional requirements based on genetic tests are also tagged with other health data to generate more comprehensive personalized nutritional product recommendations.
[0077] The method and device for providing personalized nutritional product recommendations disclosed in the present invention can generate a user portrait based on static tags indicating the user's health status and dynamic tags indicating the user's behavior, and based on the user portrait and the nutritional product knowledge graph, conveniently and accurately provide the user with personalized nutritional product recommendations. BRIEF DESCRIPTION OF THE DRAWINGS
[0078] The details of one or more implementations are set forth in the accompanying appendices, drawings, and the description that follows. Other features will be apparent from the description and drawings, and from the claims.
[0079] Figure 1 A schematic diagram of a method for providing personalized nutritional product recommendations according to a first embodiment of the present disclosure is shown.
[0080] Figure 2 A flowchart of a method for providing personalized nutritional product recommendations according to a first embodiment of the present disclosure is shown.
[0081] Figure 3 A schematic diagram of generating dynamic labels is shown.
[0082] Figure 4A schematic diagram showing a knowledge graph for nutritional products.
[0083] Figure 5 An example of a ranking model is shown.
[0084] Figure 6 A block diagram showing an apparatus for providing personalized nutritional product recommendations according to a second embodiment of the present disclosure.
[0085] Figure 7 A block diagram of a storage medium according to a third embodiment of the present disclosure is shown. DETAILED DESCRIPTION
[0086] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.
[0087] The terms used in this specification are those general terms currently widely used in the art in consideration of the functions of the present disclosure, but these terms may vary according to the intention of a person of ordinary skill in the art, precedents, or new technologies in the art. In addition, specific terms may be selected by the applicant, and in this case, their detailed meanings will be described in the detailed description of the present disclosure. Therefore, the terms used in the specification should not be understood as simple names, but rather as an overall description based on the meaning of the terms and the present disclosure.
[0088] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.
[0089] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. At the same time, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.
[0090] <First Embodiment>
[0091] Next, we will refer to Figure 1 A method for providing personalized nutritional product recommendations according to a first embodiment of the present invention is briefly described.
[0092] The method for providing personalized nutritional product recommendations according to the first embodiment of the present invention can be implemented, for example, through a WeChat applet or an application APP.
[0093] After the WeChat mini program is launched, users can register in the mini program and fill in user information including basic information such as gender and age. Then, users can conduct various evaluations or upload various evaluation data in the WeChat mini program.
[0094] like Figure 1 As shown, users will generate various labels (for example, A / D / E labels, etc.) in each evaluation, and static labels (with higher weights) for users will be generated based on these labels.
[0095] After a user has accumulated a certain amount of search records, question and answer records, and browsing records, user dynamic tags (for example, search tags, question and answer tags, short-term user tags, medium-term user tags, long-term user tags, etc.) will be generated regularly according to the tags of these records (weights decrease in sequence).
[0096] Use user static tags and user dynamic tags to form user portraits.
[0097] Items are recalled based on the physical state corresponding to the user's static tags and the behavioral records of users across the platform. Items that users may need, collect, share, and browse are preliminarily selected to form a user-item list. At the same time, multi-channel recall data is generated based on the interaction records of all users, such as popular items, the latest items, etc.
[0098] Use user portraits and item features as inputs to the sorting model to sort the user-item list and obtain a sorting result of items that the user may want to see / buy. The item sorting result is personalized and suitable for the user.
[0099] The following will refer to Figure 2 A method for providing personalized nutritional product recommendations according to a first embodiment of the present disclosure is described in detail.
[0100] Figure 2 FIG. 1 is a flow chart showing a method for providing personalized nutritional product recommendations according to a first embodiment of the present disclosure. Figure 2 As shown, methods for providing personalized nutritional product recommendations include:
[0101] S201: Acquire multiple dimensions of physiological data information of the user;
[0102] S202: generating a plurality of static tags indicating the health status of the user according to the physiological data information of the plurality of dimensions;
[0103] S203: Acquire multiple behavior data of the user;
[0104] S204: generating a plurality of dynamic tags indicating user preferences according to the plurality of behavior data;
[0105] S205: Generate a user portrait based on the static tag and the dynamic tag;
[0106] S206: Based on the multiple dynamic tags and the knowledge graph based on the nutritional products, generate multiple recall lists indicating the correspondence between users and different nutritional products through the trained recall model, and construct a recall fusion list that fuses the multiple recall lists;
[0107] S207: constructing a feature vector for a ranking model using the recall fusion list and the user portrait, and inputting the calculated feature vector into the ranking model;
[0108] S208: Sort the recalled fusion list according to the scores calculated by the sorting model on the input feature vectors to determine recommendations of personalized nutritional products suitable for the user.
[0109] Specifically, the multiple dimensions of physiological data information obtained in step S201 include one or more of the user's physical examination report data, the user's physiological data received from the wearable device, the user's self-reported nutritional needs data, and the user's genetic test report data.
[0110] For example, the user can fill in the questionnaire provided in the mini program and provide the nutritional needs that he or she is concerned about. An example of the content of the questionnaire is shown below.
[0111] A001Please select 1-4 nutritional needs that you are most concerned about: [the number of choices should not be more than 4 and not less than 1]
[0112] A. Skin, hair, nails
[0113] B. Emotional stress
[0114] C. Sleep quality
[0115] D. Resistance
[0116] E. Gastrointestinal health
[0117] F. Liver Health
[0118] G. Sports and body shape management
[0119] H. Metabolic activity and fatigue
[0120] I. Eyes
[0121] J. Exercise and bone and joint health
[0122] K. Cardiovascular health
[0123] A002 Age
[0124] A003 Gender (male / female)
[0125] A004 Weight (kg)
[0126] A005 Height (cm)
[0127] A006 (If female) Are you currently
[0128] A. Preparing for pregnancy
[0129] B. Pregnancy
[0130] C.Breastfeeding
[0131] D. Postpartum recovery period (after stopping breastfeeding)
[0132] None. None of the above (single choice)
[0133] A007. Are you taking medication (yes / no)?
[0134] A008. Have you had the following symptoms recently? [Multiple choices are allowed, only one choice is needed for the last one]
[0135] A. Night sweats
[0136] B. Hot flashes
[0137] C. Mood swings
[0138] none. None of the above
[0139] A009. Are you taking the following medicines? [Multiple choices are allowed, the last choice is single]
[0140] A. Proton pump inhibitors
[0141] B. Thiazolidinediones hypoglycemic drugs
[0142] C. Glucocorticoid drugs
[0143] D.Thyroxine
[0144] E. Anti-epileptic drugs
[0145] none. None of the above
[0146] A010. How is your mental state recently? [Multiple choices are allowed, the last choice is single]
[0147] A. Easy to get tired
[0148] B.High pressure
[0149] C. Lack of energy
[0150] D. Difficulty concentrating
[0151] E. Feeling memory deterioration
[0152] none. None of the above
[0153] Only some questionnaire examples are shown above. Other questionnaire questions may include: daily sleep duration, sleep problems, eye problems, hair, skin, nail problems, emotional state, stress level, various physical discomfort symptoms, diet, living habits, exercise intensity, bone problems, blood sugar, blood, blood lipid diseases, recent nutritional deficiency diagnosis, etc.
[0154] In addition, users can also upload their own physical examination reports in the mini program. Physical examination reports provide reference ranges and test index values for blood pressure, triglycerides (TG), total cholesterol (TC), fasting blood sugar, alanine aminotransferase (ALT), hemoglobin (HGB), carotid artery color Doppler ultrasound, etc. The following are some examples of submitted reports.
[0155]
[0156] In addition, users can also configure their wearable devices (such as smart watches) to connect with WeChat applet to obtain user exercise, sleep and other data. In some cases, smart watches can also be used to obtain user blood sugar, blood pressure, heart rate, stress and other data. The following are some examples of exercise and sleep data.
[0157]
[0158] Finally, users can also upload their genetic test results to the WeChat applet. An example of a genetic test indicator is shown below.
[0159]
[0160]
[0161] In step S202, after uploading the user's data, various static tags of the user can be generated using these data. As shown above, the user fills out a nutritional needs questionnaire, and a tag A is generated based on the questionnaire. The user uploads a physical examination report, and a tag B is generated based on the results of the submitted report. A tag C is generated based on the data obtained by the wearable device. The user uploads a genetic test report, and a tag D is generated based on the results of the genetic test report.
[0162] Specifically, a first type of tag with a first weight is generated according to the user's physical examination report data, and the first type of tag includes multiple tags indicating different physiological health states of the user. For example, the user uploads a physical examination report, and a B tag is generated according to the result of submitting the report.
[0163] A second type of tag with a second weight is generated according to the physiological data of the user received from the wearable device, wherein the second type of tag includes a plurality of tags indicating different physiological activity states of the user. For example, a C tag is generated according to the data obtained by the wearable device.
[0164] A third type of label with a third weight is generated according to the nutritional requirement data reported by the user himself, and the third type of label includes multiple labels indicating different mental states and nutritional requirements reported by the user himself. For example, the user fills out a nutritional requirement questionnaire, and a label A is generated according to the questionnaire.
[0165] A fourth type of tag with a fourth weight is generated according to the user's gene test report data, and the fourth type of tag includes a plurality of tags indicating different long-term nutritional needs of the user. For example, the user uploads a gene test report, and a D tag is generated according to the result of the gene test report.
[0166] A plurality of static tags indicating the health status of the user are generated based on the user's basic information, the first type of tags, the second type of tags, the third type of tags, and the fourth type of tags. For example, according to the algorithm rules of meta-health, an E tag is generated based on the user's basic information, part or all of the A tag, the B tag, the C tag, and the D tag.
[0167] In this embodiment, these tags have different weights, and the weight values are allocated in the following order: physical examination index > exercise and sleep data > questionnaire > genetic testing (long-term basic needs). Therefore, the first weight, the second weight, the third weight and the fourth weight have decreasing values in sequence.
[0168] In one embodiment, each tag represents a judgment of the user's physical condition, which is obtained from the contents of the questionnaire and the physical examination report. The tag result is 0 / 1 / 2, where 0 means that it is impossible to judge that the user and the tag match, 1 means that the user is more likely to match the tag, and 2 means that the user and the tag are confirmed to match.
[0169] The table below lists all possible answers that can be labeled. For example, for the E006 label, if the user selects option a of question A011 in the questionnaire, we believe that the user can be judged to have insomnia from the questionnaire. Most of the labels hit by the questionnaire answer will have the label value set to 2, and in rare cases it will be 1 (such as E012).
[0170] By default, as long as the judgment conditions are met, the label value is 2. If the label value is 1, a description of => 1 will be added at the end. If there are multiple judgment criteria that can be used to label (such as labels from questionnaire answers and physical examination reports at the same time), the label result takes the maximum value.
[0171] Some examples of E tags are shown below.
[0172]
[0173]
[0174]
[0175] The user's basic information, A label, B label, C label, D label, and E label are combined into a user static label. The following is an example of a static label.
[0176]
[0177] In step S203, multiple behavior data of the user may be obtained. The multiple behavior data of the user include user browsing records, user search records, user question and answer records, and corresponding time records. For example, after a user has used a program for a period of time, some user interaction records will be generated. The interaction records include "browsing products", "search records", "viewing question and answer pairs", etc. These interaction records are the user's behavior data.
[0178] Then, in step S204, a plurality of dynamic tags indicating the user's preferences may be generated according to the plurality of behavior data.
[0179] Generating a plurality of dynamic tags indicating the user's preferences according to the plurality of behavior data comprises:
[0180] Generate browsing tags, search tags and question-and-answer tags based on user browsing records, user search records and user question-and-answer records of multiple users, wherein the tags include three types: entity, keyword and classification;
[0181] Building multiple collections of browsing tags, search tags, and question-and-answer tags of multiple users within a predetermined time period based on the user as a unit;
[0182] Calculating a score for each user tag in the plurality of sets using a term frequency-inverse document frequency (TF-IDF) algorithm;
[0183] Generate the final score of each user tag based on the score of each user tag, the time decay of the interaction, the weight of the interaction type, and the weight of the tag type;
[0184] A predetermined number of user tags with the highest final scores are selected as dynamic tags of the users.
[0185] Specifically, for example, the embodiment of the present application will generate dynamic tags such as "search tags", "question and answer tags", "short-term user tags (7 days)", "medium-term user tags (30 days)", and "long-term user tags (365 days)" based on user interaction records.
[0186] Figure 3 The generation process of dynamic labels is shown in Figure 2. Figure 3 As shown, the steps for generating dynamic tags are as follows:
[0187] a. Construct a collection of articles / product tags recorded by users on the entire platform within a predetermined time window, including searches / question-answer pairs / browsing (7 days) / browsing (30 days) / browsing (365 days), by user, for example
[0188] User A: [label 1, label 2, label 5, label 9]
[0189] User B: [label 3, label 4, label 7]
[0190] User C: [label 2, label 6, label 8, label 10]
[0191] In this embodiment, tags are divided into three types of tags, for example, "entity", "keyword" and "classification".
[0192] b. Use the term frequency-inverse document frequency (TF-IDF) algorithm to calculate the score of each user tag.
[0193] c. Calculate the score of each user tag by combining tf-idf score, time decay, interaction type (view, like, favorite), and tag type (entity, keyword, category), and use the tags ranked in the top 10 or top 5 as the user's dynamic tags.
[0194] In this embodiment, for example, the tag 1 score of user A=tf-idf score*time decay of generating interaction*interaction type weight*tag type weight.
[0195] Below is an example of a user dynamic tag:
[0196]
[0197]
[0198] In step S205, a user portrait may be generated based on the generated static tags and dynamic tags. The user portrait is a pattern composed of static tags + dynamic tags + user information in the form of tags.
[0199] Then, in step S206, based on the multiple dynamic tags and the knowledge graph based on nutritional products, multiple recall lists indicating the correspondence between users and different nutritional products are generated through the trained recall model, and a recall fusion list that fuses the multiple recall lists is constructed.
[0200] Specifically, the recall model may include multiple recall sub-models for recalling various nutritional products related to the user. For example, the recall model may include a SVD recall sub-model based on a singular value decomposition model (SVD), an update recall sub-model, a hot recall sub-model, and a graph recall sub-model.
[0201] The SVD recall submodel is used to generate an SVD recall list indicating the products that the user is most likely to purchase. The rating matrix used in the SVD model is composed of users and items. The following table is an example.
[0202] User 1 User 2 User 3 User 4 User 5 Item A 4 3 2 Item B 5 1 Item C 2 4
[0203] Decompose the given rating matrix S into the product of three matrices, where U and V are called left and right singular vectors, which can be understood as the user matrix and item factor matrix in this embodiment. The values on the diagonal are called singular values; S is an n*m matrix, U is an n*n matrix, Q is an n*m matrix, and V is an m*m matrix; in the following formula, k singular values can be used to approximately replace the R matrix, because the sum of the first 1% of singular values accounts for more than 99% of the sum of all singular values, because except for the middle singular values, the rest of the singular values are basically 0. The formula is as follows:
[0204] S n*m =U n*n *Q n*m *V m*n S n*m ≈U n*k *Q k*k *V k*n Formula (1)
[0205] Then we need to transform the U and V matrices to get the user factor matrix C and the item factor matrix P, and take the square root of the singular matrix Q and multiply it to the U and V matrices respectively, just like taking the square root of the singular matrix Q and then multiplying it. We get the user factor matrix C and the item factor matrix P as follows:
[0206]
[0207] Finally, the rating prediction of user t for item i is the t-th row of the user factor matrix multiplied by the i-th column of the item factor matrix (i.e., the transpose of the i-th row of the item factor matrix), as shown below:
[0208]
[0209] The update recall submodel is used to generate an update recall list of products indicating the most recently updated ones.
[0210] The hot recall sub-model is used to generate a hot recall list of popular products based on the click volume and time decay of the products.
[0211] For example, the update recall sub-model is sorted in reverse order according to the release time of products / articles and the top n products / articles are selected as the hot recall objects. The hot recall formula is, for example, the following formula:
[0212]
[0213] As shown in the above formula, heat is the heat of the article / product, n is the total number of interactions, t i is the time when the i-th interaction occurs, r count is the amount of user interaction (if only browsing, then r count 1, if there is browsing or collection, then r count For 2)
[0214] The graph recall sub-model is used to query products associated with the tags from the knowledge graph based on nutritional products according to the user's multiple tags, and generate a graph recall list of personalized products.
[0215] Figure 4 The knowledge graph based on nutritional products is shown. The knowledge graph based on nutritional products adopts triples in the form of <entity, relationship, entity>.
[0216] The nutritional knowledge graph includes the associations between multiple nutritional elements and at least one health state. One association in the nutritional knowledge graph represents the effect of a nutritional element on a health state. The effect of a nutritional element on a nutritional indicator may include positive effect, negative effect or no effect, wherein a positive effect means that the nutritional element contributes to the health state and has a positive effect on the health state; a negative effect means that the nutritional element is harmful to the health state and has a negative effect on the health state; and no effect means that the nutritional element has no effect on the health state.
[0217] like Figure 4 As shown, for example, soy isoflavones contribute to bone health, L-theanine helps relieve stress-induced tension, vitamin E acts as an antioxidant to protect cells from free radicals, vitamin D has an adverse effect on high blood calcium, and so on.
[0218] The graph recall sub-model can retrieve multiple corresponding nutritional elements from the graph according to multiple tags of the user (such as E tags). Then, multiple nutritional products corresponding to the multiple nutritional elements are retrieved in the database. Finally, based on the search results, a personalized user-nutritional product graph recall list specific to each user is formed.
[0219] The recall model may further include a fusion sub-model for fusing the SVD recall list, the update recall list, the hot recall list and the atlas recall list according to a predetermined fusion rule to generate a recall fusion list.
[0220] An example of recall fusion rules is shown in the table below. Assume that there are 3 recall strategies and 5 products. Product A is recalled by recall strategies 1, 2, and 3, and product C is also recalled by the three recall strategies. The score of products A and C is 3, and product B is only recalled by recall strategies 2 and 3 with a score of 2. Then, when recall fusion is performed, if the recall quantity is 3, only products A, B, and C will be recalled.
[0221] Recall Strategy 1 Recall Strategy 2 Recall Strategy 3 Total score A √ √ √ 3 B √ √ 2 C √ √ √ 3 D √ 1 E √ 1
[0222] In step S207, a feature vector for a ranking model is constructed using the recall fusion list and the user portrait, and the calculated feature vector is input into the ranking model.
[0223] Specifically, constructing a feature vector for a ranking model using the recall fusion list and the user portrait includes:
[0224] Perform feature engineering on each information field in the user portrait. Because the input data of the ranking model has a specific format, it is necessary to perform feature engineering on each information field in the user portrait in order to obtain formatted information.
[0225] An example of a user portrait is shown below.
[0226]
[0227]
[0228] Model input after feature processing.
[0229]
[0230]
[0231] Then, according to the weights of the various features, a weight assignment process is performed to generate a feature vector for a sorting model, wherein the weights in the static labels are greater than the weights in the dynamic labels.
[0232] Model input after weight assignment
[0233]
[0234] In step S208, the multiple recall lists included in the recall fusion list are sorted according to the scores calculated by the sorting model for the input feature vectors to determine the recommendation of personalized nutritional products suitable for the user.
[0235] Specifically, the ranking model calculates a score indicating the user's interest in the product based on the input feature vector, and sorts the multiple recall lists according to the value of the score.
[0236] Finally, the ranking model determines the nutritional products in the recall list with the highest score as the recommendation of personalized nutritional products suitable for the user. In other words, the ranking model scores the user's "interest level" in the product based on the input, and the recall list with higher scores will be presented to the user first.
[0237] The sorting model is based on the DeepFM model, for example. The DeepFM model combines the advantages of wide and deep models, jointly training the FM model and the DNN model to simultaneously learn low-order feature combinations and high-order feature combinations. In addition, the Deep component and the FM component of the DeepFM model share data input from the Embedding layer.
[0238] Figure 5 The system block diagram of DeepFM is shown. DeepFM consists of two parts, the FM part on the left and the DNN part on the right. The two parts share the same input. For a given feature i, wi is used to represent the importance of the first-order feature, and the latent vector Vi of feature ii is used to represent the interaction with other features. In the FM part, Vi is used to characterize the second-order features, and in the neural network part it is used to construct high-order features. For the current model, all parameters participate in the training together. The prediction result of DeepFM can be written as y∈(0,1) is the predicted CTR, is the result obtained by the FM part, and is the result of the DNN part.
[0239] Specifically, DeepFM consists of two parts, the FM part on the left and the DNN part on the right. Its prediction formula can be written as
[0240]
[0241] where y FM for:
[0242]
[0243] where w∈R d, x is the input feature,<w,x> It is an addition unit used to calculate the first-order features. For a given feature i,v i Used to represent the interaction with other features and second-order features.
[0244] where y DNN for:
[0245] y DNN =σ(W |H|+1 ·a H +b |H|+1 )
[0246] Where |H| is the number of hidden layers of the DNN neural network, W is the weight of the DNN neural network, b is the bias term of the DNN neural network, and a is the input feature.
[0247] The ranking model calculates a score indicating the degree of association between the product and the user based on the input feature vector, and sorts the multiple recall lists according to the value of the score. Finally, the ranking model can determine the nutritional product in the recall list with the highest score as a recommendation of the personalized nutritional product suitable for the user.
[0248] In a preferred embodiment, based on the user portrait, the personalized nutritional products recommended to the user include a combination of nutritional products of different types and quantities corresponding to the user's health status. In this way, the user can directly understand the personalized products he needs without having to go to the hospital to see a doctor.
[0249] The following shows examples of combined products for different user profiles.
[0250]
[0251]
[0252] In a preferred embodiment, the user's health status can also be determined based on the user's static tags, associated nutritional products can be obtained based on the user's health status and the knowledge graph of nutritional products, and recommendations of related nutritional products suitable for the user can be determined based on the obtained associated nutritional products.
[0253] After obtaining the recommendation of the personalized nutritional product for the user, the recommendation of the personalized nutritional product and / or the recommendation of the related nutritional product may also be displayed to the user through the WeChat mini program.
[0254] In a preferred embodiment, the user's health status can also be determined based on the user's static tags. The determined health status and health advice corresponding to the health status are displayed to the user. For example, the determined health status and health advice corresponding to the health status can be displayed to the user through a WeChat applet.
[0255] In a preferred embodiment, the user's nutritional needs can also be determined based on the user's static tags, associated nutritional products can be obtained based on the user's nutritional needs and the knowledge graph of nutritional products, and recommendations of related nutritional products suitable for the user can be determined based on the associated nutritional products obtained.
[0256] For example, the user's nutritional needs and / or recommendations of the personalized nutritional products can be displayed to the user through the WeChat mini program.
[0257] In a preferred embodiment, the above-described method of determining a user's nutritional needs and recommending related nutritional products suitable for the user and / or determining a user's health status and health advice corresponding to the health status is performed, for example, in a remote server.
[0258] The remote server can then send the determined nutritional needs of the user and recommendations of related nutritional products suitable for the user and / or determine the health status of the user and health advice corresponding to the health status to the user's electronic device, and then display them on the user's electronic device.
[0259] According to the method for providing personalized nutritional product recommendations in accordance with the first embodiment of the present disclosure, it is possible to analyze the user's data with the help of data modeling, machine learning, and medical knowledge, thereby providing personalized insights and recommendations for the user. In addition, according to the method for providing personalized nutritional product recommendations in accordance with the present disclosure, it is also possible to obtain recommended nutritional requirements from genetic tests, rather than directly investigating the original genetic test data. In addition, nutritional requirements based on genetic tests are also tagged with other health data to generate more comprehensive personalized nutritional product recommendations.
[0260] Therefore, the method for providing personalized nutritional product recommendations according to the first embodiment of the present disclosure can generate a user portrait based on static tags indicating the user's health status and dynamic tags indicating the user's behavior, and based on the user portrait and the nutritional product knowledge graph, conveniently and accurately provide the user with personalized nutritional product recommendations.
[0261] <Second Embodiment>
[0262] Next, we will refer to Figure 6 An apparatus for providing personalized nutritional product recommendations according to a second embodiment of the present invention is described.
[0263] like Figure 6 As shown, the apparatus 700 for providing personalized nutritional product recommendations according to the second embodiment includes a memory 601 and a processor 602 .
[0264] The memory 601 stores computer programs.
[0265] The processor 602 can execute the computer program in the memory 601. When the processor executes the program, it can perform:
[0266] Obtain multiple dimensions of physiological data information of users;
[0267] Generating a plurality of static tags indicating the health status of the user according to the physiological data information of the plurality of dimensions;
[0268] Obtain multiple behavioral data of users;
[0269] generating a plurality of dynamic tags indicating preferences of the user according to the plurality of behavior data;
[0270] Generate a user portrait based on the static tags and the dynamic tags;
[0271] Based on the multiple dynamic tags and the knowledge graph based on the nutritional products, multiple recall lists indicating the correspondence between users and different nutritional products are generated through the trained recall model, and a recall fusion list that fuses the multiple recall lists is constructed;
[0272] constructing a feature vector for a ranking model using the recall fusion list and the user portrait, and inputting the calculated feature vector into the ranking model; and
[0273] The recall fusion list is ranked according to the scores calculated by the ranking model on the input feature vectors to determine the recommendation of personalized nutritional products suitable for the user.
[0274] When such a device 600 executes a program, the method described in the first embodiment above is executed, and its detailed description is omitted here.
[0275] The apparatus 600 is, for example, an electronic device with computing capabilities, such as a mobile terminal, a desktop computer, a notebook computer, a server, or a portable computing device.
[0276] The device 600 may also include a display that can display recommendations of personalized nutritional products suitable for the user to the user.
[0277] The apparatus 600 may also be a system including multiple separate components. For example, a processing component is deployed on or implemented as a remote server, and a display and / or another processing component is deployed on a local mobile terminal. The remote server may send processed information to the local mobile terminal for display.
[0278] Therefore, the device for providing personalized nutritional product recommendations using the second embodiment of the present disclosure can generate a user portrait based on static tags indicating the user's health status and dynamic tags indicating the user's behavior, and based on the user portrait and the nutritional product knowledge graph, conveniently and accurately provide the user with personalized nutritional product recommendations.
[0279] Based on the above embodiments, an embodiment of the present disclosure further provides a computer-readable storage medium. Figure 7 1000 is a schematic diagram of a storage medium according to an embodiment of the present disclosure. Figure 7 As shown, the computer-readable storage medium 1000 stores computer-executable instructions 1001. When the computer-executable instructions 1001 are executed by a processor, the method for providing personalized nutritional product recommendations in the above embodiment can be implemented.
[0280] The computer readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.
[0281] The embodiments of the present disclosure also provide a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device implements the apparatus for providing personalized nutritional product recommendations in the above embodiments.
[0282] Those skilled in the art will appreciate that the contents disclosed in this disclosure may be subject to various modifications and improvements. For example, the various devices or components described above may be implemented by hardware, or by software, firmware, or a combination of some or all of the three.
[0283] In addition, although the present disclosure makes various references to certain units in the system according to embodiments of the present disclosure, any number of different units may be used and run on the client and / or server. The units are illustrative only, and different aspects of the systems and methods may use different units.
[0284] Those skilled in the art will appreciate that all or part of the steps in the above method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk. Optionally, all or part of the steps in the above embodiment can also be implemented using one or more integrated circuits. Accordingly, each module / unit in the above embodiment can be implemented in the form of hardware or in the form of a software functional module. The present disclosure is not limited to any particular form of combination of hardware and software.
[0285] In summary, it should be appreciated that although specific embodiments of the disclosed technology have been described herein for purposes of illustration, various modifications may be made without departing from the scope of the invention. Accordingly, the disclosed technology is not to be limited except as set forth in the appended claims.
[0286] The implementation of the subject matter and functional operations described in this patent document can be implemented in various systems, digital electronic circuits, or in computer software, firmware or hardware (including the structures disclosed in the specification and their structural equivalents), or in one or more combinations thereof. The implementation of the subject matter described in this specification can be implemented as one or more computer program products, that is, one or more modules of computer program instructions encoded on a tangible and non-temporary computer-readable medium, for execution by a data processing device or control of the operation of the data processing device. The computer-readable medium can be a machine-readable storage device, a machine-readable storage substrate, a memory device, a composition of a substance that affects a machine-readable propagation signal, or one or more combinations thereof. The term "data processing unit" or "data processing device" covers all devices, devices and machines for processing data, including, for example, a programmable processor, a calculator, or multiple processors or computers. In addition to hardware, the device may also include code to create an execution environment for the mentioned computer program, for example, code constituting processor firmware, a protocol stack, a database management system, an operating system, or one or more combinations thereof.
[0287] A computer program (also referred to as a program, software, software disclosure, script or code) can be written in any form of programming language, including assembly or interpretation language, and it can be deployed in any form, including as an independent program or module, component, subroutine or other unit suitable for use in a computing environment. A computer program does not necessarily correspond to a file in a file system. A program can be stored in a portion of a file with other programs or data (e.g., stored in one or more scripts in a markup language file), in a single file dedicated to the program mentioned, or in multiple coordinated files (e.g., files storing one or more modules, subroutines or partial codes). A computer program can be deployed to execute on one or more computers, which are located at one site or distributed across multiple sites and interconnected by a communication network.
[0288] The processors and logic flows described in this specification can be performed by one or more programmable processors executing one or more computer programs to perform functions by operating input data and generating output. The processors and logic flows can also be performed by special purpose logic circuits, and the device can also be implemented as special purpose logic circuits, such as FPGAs (field programmable gate arrays) or ASICs (application-specific integrated circuits).
[0289] Processors suitable for executing computer programs include, for example, general and special purpose microprocessors, and any one or more processors of any kind of digital computer. Typically, the processor will receive instructions or data from a read-only memory or a random access memory or both. The essential elements of a computer are a processor for executing instructions and one or more memory devices for storing instructions and data. Typically, a computer will also include one or more large-capacity storage devices for storing data, such as a disk, a magneto-optical disk, or an optical disk, or be operatively coupled to receive data from one or more large-capacity storage devices for storing data or to transmit data to it. However, a computer does not necessarily have these devices. Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media, and memory devices, including, for example, semiconductor memory devices (e.g., EPROM, EEPROM, and flash memory devices). The processor and memory can be supplemented by a dedicated logic circuit, or incorporated into a dedicated logic circuit.
[0290] The specification and drawings are to be regarded as exemplary only, wherein exemplary refers to one example. As used herein, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. Furthermore, the use of "or" is intended to include "and / or" as well, unless the context clearly indicates otherwise.
[0291] Although the present invention document contains many details, these should not be interpreted as limitations on the scope of any invention or claim, but should be used as descriptions of features specific to a particular embodiment of a particular invention. Certain features described in this patent document in the context of a separate embodiment can be implemented in a single embodiment in combination. Conversely, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination. In addition, although certain features may be described above as working in certain combinations and even initially claimed for protection, one or more features from the claimed combination may be removed from the combination in some cases, and the claimed combination may involve a sub-combination or a variation of the sub-combination.
[0292] Similarly, although operations are described in a particular order in the drawings, this should not be understood as requiring that such operations be performed in the particular order or sequence shown, or that all of the operations shown be performed to achieve the desired results. In addition, the division of various system components in the embodiments described in this patent document should not be understood as requiring such division in all embodiments.
[0293] Only a few implementations and examples are described, and other implementations, enhancements, and variations may be made based on what is described and illustrated in this patent document.
Claims
1. A computer-implemented method for providing personalized nutritional product recommendations, comprising: Obtain multiple dimensions of physiological data information of users; generating a plurality of static tags indicating the health status of the user according to the physiological data information of the plurality of dimensions; Obtain multiple behavioral data of users; generating a plurality of dynamic tags indicating preferences of the user according to the plurality of behavior data; Generate a user portrait based on the static tags and the dynamic tags; Based on the multiple dynamic tags and the knowledge graph based on the nutritional products, multiple recall lists indicating the correspondence between users and different nutritional products are generated through the trained recall model, and a recall fusion list that fuses the multiple recall lists is constructed; constructing a feature vector for a ranking model using the recall fusion list and the user portrait, and inputting the calculated feature vector into the ranking model; as well as The recall fusion list is ranked according to the scores calculated by the ranking model on the input feature vectors to determine the recommendation of personalized nutritional products suitable for the user.
2. The method as claimed in claim 1, wherein the physiological data information of multiple dimensions includes one or more of the user's physical examination report data, the user's physiological data received from a wearable device, the user's self-reported nutritional needs data, and the user's genetic test report data.
3. The method of claim 2, wherein generating a plurality of static tags indicating the physiological state of the user according to the physiological data information of the plurality of dimensions comprises: generating a first type of label with a first weight according to the user's physical examination report data, wherein the first type of label includes a plurality of labels indicating different physiological health states of the user; generating a second type of tag with a second weight according to the physiological data of the user received from the wearable device, wherein the second type of tag includes a plurality of tags indicating different physiological activity states of the user; generating a third type of tags with a third weight according to the nutritional requirement data reported by the user himself, wherein the third type of tags includes a plurality of tags indicating different mental states and nutritional requirements reported by the user himself; generating a fourth type of label with a fourth weight according to the user's genetic test report data, the fourth type of label including a plurality of labels indicating different long-term nutritional needs of the user; as well as Generate multiple static tags indicating the health status of the user according to the user basic information, the first type of tags, the second type of tags, the third type of tags and the fourth type of tags, The first weight, the second weight, the third weight and the fourth weight have decreasing values in sequence.
4. A method as described in any one of claims 1-3, wherein the user's multiple behavior data include user browsing records, user search records, user question and answer records, and corresponding time records.
5. The method of claim 4, wherein: Generating a plurality of dynamic tags indicating the user's preferences according to the plurality of behavior data includes: Generate browsing tags, search tags and question-and-answer tags based on user browsing records, user search records and user question-and-answer records of multiple users, wherein the tags include three types: entity, keyword and classification; Building multiple collections of browsing tags, search tags, and question-and-answer tags of multiple users within a predetermined time period based on the user as a unit; Calculating a score for each user tag in the plurality of sets using a term frequency-inverse document frequency (TF-IDF) algorithm; Generate the final score of each user tag based on the score of each user tag, the time decay of the interaction, the weight of the interaction type, and the weight of the tag type; A predetermined number of user tags with the highest final scores are selected as dynamic tags of the users.
6. The method of claim 5, wherein the recall model comprises: An SVD recall sub-model based on a singular value decomposition (SVD) model, for generating an SVD recall list indicating the products that the user is most likely to purchase; an update recall sub-model for generating an update recall list indicating the most recently updated products; The hot recall sub-model is used to generate a hot recall list of popular products based on the product's click volume and time decay; A graph recall sub-model, for querying products associated with the tags from the nutritional product-based knowledge graph according to multiple tags of the user, and generating a graph recall list of personalized products; as well as The fusion sub-model is used to fuse the SVD recall list, the update recall list, the hot recall list and the atlas recall list according to a predetermined fusion rule to generate a recall fusion list.
7. The method of claim 6, wherein constructing a feature vector for a ranking model using the recall fusion list and the user profile comprises: Perform feature engineering on each information field in the user portrait; as well as According to the weights of the various features, a weight assignment process is performed to generate a feature vector for a sorting model, wherein the weights in the static labels are greater than the weights in the dynamic labels.
8. The method of claim 7, wherein ranking the recall fusion list according to the scores calculated by the ranking model on the input feature vectors to determine the recommendation of personalized nutritional products suitable for the user comprises: The ranking model calculates a score indicating the user's interest in the product based on the input feature vector; sorting the multiple recall lists contained in the recall fusion list according to the values of the scores; as well as The nutritional product in the recall list with the highest score is determined as a recommendation of a personalized nutritional product suitable for the user.
9. The method of claim 1, wherein: The personalized nutritional product includes a combination product unit of a plurality of different types and different quantities of nutritional products corresponding to the health status of the user.
10. The method of claim 1, further comprising: Determine the user's health status based on the user's static tags; Obtaining related nutritional products based on the user's health status and the knowledge graph of nutritional products; Based on the obtained associated nutritional products, a recommendation of related nutritional products suitable for the user is determined.
11. The method of claim 10, further comprising: Displaying the personalized nutritional product recommendation and / or the related nutritional product recommendation to the user; or sending the personalized nutritional product recommendation and / or the related nutritional product recommendation to another electronic device, and The recommendation of the personalized nutritional product and / or the recommendation of the related nutritional product are displayed on another electronic device.
12. The method of claim 1, further comprising: Determine the user's health status based on the user's static tags; displaying the determined health status and health advice corresponding to the health status to the user; or sending the determined health status and health advice corresponding to the health status to another electronic device, and The determined health status and health advice corresponding to the health status are displayed on another electronic device.
13. The method of claim 1, further comprising: Determine the user's nutritional needs based on the user's static tags; Obtaining related nutritional products based on the user's nutritional needs and the knowledge graph of nutritional products; Based on the obtained associated nutritional products, a recommendation of related nutritional products suitable for the user is determined.
14. The method of claim 13, further comprising: Displaying to a user the nutritional needs of the user and / or recommendations of the personalized nutritional products; or sending the personalized nutritional product recommendation and / or the related nutritional product recommendation to another electronic device, and The recommendation of the personalized nutritional product and / or the recommendation of the related nutritional product are displayed on another electronic device.
15. A device for providing personalized nutritional product recommendations, comprising: a memory having a computer program stored thereon; as well as The processor is configured to, when executing the computer program, perform: Obtain multiple dimensions of physiological data information of users; generating a plurality of static tags indicating the health status of the user according to the physiological data information of the plurality of dimensions; Obtain multiple behavioral data of users; generating a plurality of dynamic tags indicating preferences of the user according to the plurality of behavior data; Generate a user portrait based on the static tags and the dynamic tags; Based on the multiple dynamic tags and the knowledge graph based on the nutritional products, multiple recall lists indicating the correspondence between users and different nutritional products are generated through the trained recall model, and a recall fusion list that fuses the multiple recall lists is constructed; constructing a feature vector for a ranking model using the recall fusion list and the user portrait, and inputting the calculated feature vector into the ranking model; as well as The recall fusion list is ranked according to the scores calculated by the ranking model on the input feature vectors to determine the recommendation of personalized nutritional products suitable for the user.
16. A non-volatile storage medium having a computer program stored thereon, wherein when the program is executed by a computer, the method according to any one of claims 1 to 14 is performed.
