Large model modeling method of delicious food big data

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a big model modeling method for delicious food big data, which comprises the following steps: S1, acquiring old user historical data, store delicious food data, new user basic data, social platform data and shopping platform data; s2, processing the basic data of the new user and platform data related to the new user; s3, analyzing the activity data of the new user on the social platform and the shopping platform; according to the method, the data of the new users and the activity data of the new users on other platforms are obtained, the data are analyzed and then compared with the historical data of the old users, after similar features are obtained, the new users are classified, the preferences of the new users are predicted based on the preferences of the old users on the store food, and personalized store recommendation is made. And the purchase and attention of the new user on the diet on other platforms are taken as the basis of associated recommendation, so that the accuracy of shop recommendation is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method for building a large model of food big data. Background Art

[0002] Big data refers to a collection of data that is huge in quantity, diverse in types, and generated at a fast speed. This data exceeds the storage, management, and processing capabilities of traditional database systems.

[0003] Big model modeling refers to building and training machine learning models based on big data and complex neural network structures, especially in the field of deep learning.

[0004] The patent application number is CN202010841508.X, which states in the specification that "the present invention discloses a method and system for recommending food shops. The embodiment of the present invention calculates the similarity between the representation vector of the user portrait and the representation vector of each food shop based on the acquired data of a user portrait and the acquired data of multiple food shops; the calculated similarity and the acquired user historical scoring data are input into the trained autoencoder neural network model for score prediction to obtain the score of the food shop; the food shop is sorted according to the score to obtain the food shop recommended for the user. In the embodiment of the present invention, since the data of the user portrait includes not only the user's comments but also multi-dimensional user feature data, the embodiment of the present invention can accurately recommend food shops based on multi-dimensional user feature data." Although the above technology has the advantage of accurately recommending food shops based on multi-dimensional user feature data, when facing new users without historical data, the above technology still has the problem of not being able to accurately recommend suitable shops.

[0005] In summary, developing a large model modeling method for food big data is still a key issue that needs to be urgently solved in the field of artificial intelligence technology. Summary of the invention

[0006] In order to solve the above problems, the present invention provides a large model modeling method for food big data. The present invention obtains the data of new users and the activity data of new users on other platforms, analyzes these data and compares them with the historical data of old users. After obtaining similar features, the new users are classified, and the preferences of new users are predicted based on the preferences of old users for store food, and personalized store recommendations are made. The purchase and attention of new users to food on other platforms are used as the basis for associated recommendations, thereby improving the accuracy of recommended stores. At the same time, the model obtains feedback data from new users in real time, and measures the deviation between the feedback data and the model predicted data. The model parameters are optimized and updated through the feedback data of new users, so that the model can better capture the user's preferences, gradually improve the accuracy of the present invention in predicting the preferences of new users, and further improve the accuracy of recommending stores to new users.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] The present invention provides a method for building a large model of food big data, comprising the following steps: S1, obtaining old user historical data, store food data, new user basic data, social platform data and shopping platform data;

[0009] S2. Process the basic data of the new user and the platform data related to the new user;

[0010] S3. Analyze the activity data of new users on social platforms and shopping platforms;

[0011] S4. Collaboratively filter the analyzed data with the historical data of old users and the food data of the store;

[0012] S5, obtain the activity data of new users in the model in real time;

[0013] S6. Continuously update and optimize the model through feedback data from new users.

[0014] Furthermore, in step S1, the method for obtaining old user historical data, store food data, new user basic data, social platform data and shopping platform data is:

[0015] Access the user data module through the internal API to obtain user activity data. Store food data is obtained through the merchant's registration, and can also be obtained from user comments and feedback data. New user basic data is obtained during registration, and new users are guided to choose food preferences during registration. After the user authorizes the social platform data and shopping platform data, the user's information including but not limited to social behavior, published content, shopping history, preference data and browsing data is obtained.

[0016] Furthermore, in step S2, the method for processing the basic data of the new user and the platform data related to the new user is:

[0017] Fill in the missing values ​​in the acquired new user basic data and standardize the user's numerical data. The filling formula is:

[0018] in is the estimated value of the missing value of user i in dimension j, is the global average of all user feature data, a i is the bias term of user i, representing the overall preference of the user, a j is the bias term of feature j, representing the overall preference of the feature, a k is the bias of other features k related to feature j, which is used to adjust the estimate m when calculating missing values ​​in collaborative filtering jk is the similarity weight between feature j and other features k, K is the set of all other features, y ik is the known value of feature k for user i, and the normalization formula is: Where Q ij is the standardized value of the i-th user on the j-th feature, y ij is the original data value of the i-th user on the j-th feature, m j is the mean of the jth feature, that is, the average value of all users on this feature: δ j is the standard deviation of the jth feature, that is, the degree of fluctuation of the feature: n is the total number of users. When processing multi-dimensional data, the feature matrix of the entire new user can be standardized. Let C be an n×t matrix, representing n users and t features, then the matrix formula is: Where B is the normalized matrix, m is a column vector containing the mean of each feature with a size of 1×t, and δ is a column vector containing the standard deviation of each feature with a size of 1×t, which normalizes the feature data of all users.

[0019] Furthermore, in step S3, the method for analyzing the activity data of the new user on the social platform and the shopping platform is:

[0020] The data of new users on social platforms and shopping platforms are associated and mined. After obtaining the data, the data is cleaned and formed into an event log table. The click path, dwell time and conversion rate of the new users' activity data are analyzed. The data of new users are compared and identified with the data of old users. The comparison and identification content includes but is not limited to gender, age, geographic location, and purchased food, so as to identify similar types of user groups and classify them.

[0021] Furthermore, in step S3, the method for analyzing the activity data of the new user on the social platform and the shopping platform is:

[0022] The process of analyzing the click path, dwell time, and conversion rate in comparative identification is as follows: i =[q i1 ,q i2 ,q i3 ,···q in ], where P i is the click sequence of the i-th user, q in is the nth click behavior of the user, including but not limited to product pages, activity pages and food videos; the dwell time analysis formula is: where S i The total stay time of user i, S in is the user's stay time on page n, and the conversion rate analysis formula is: Where Z i is the ratio of user i from click to actual purchase, k is the number of conversion events, and N is the total number of clicks. By analyzing the dwell time of new users on different pages and the purchase conversion rate, the preferences of new users can be identified.

[0023] Furthermore, in step S3, the method for analyzing the activity data of the new user on the social platform and the shopping platform is:

[0024] The similarity calculation formula is: Where x is a new user, y is an old user, and u x，n and u y，n is the value of feature n of user x and user y. The value range of cosine similarity is [-1, 1]. The closer the value is to 1, the more similar they are. The classification formula based on similarity is: where H is the number of clusters, Z h is the user set of the hth cluster, v i is the feature vector of the i-th user, κ h It is the center point of the hth cluster. By analyzing the activity data of new users, the new user groups can be identified and classified.

[0025] Furthermore, in step S4, the method of collaboratively filtering the analyzed data with the old user historical data and the store food data is as follows:

[0026] After analyzing new users, we can identify the similarities between new users and old users. Based on the similarities between new users and old users, and based on the old users' preferences for store food, we can predict the food stores preferred by new users. The prediction formula is:

[0027] in is the score of store i predicted for new user x, x n is the historical average rating of new user x, Z x is the set of old users similar to the new user x, sim(x, y) is the similarity between the new user x and the old user y, is the actual rating of old user y on store i, l y is the mean rating of old users y. By considering the ratings and comments of old users on the store, we can predict the ratings and comments of new users who are similar to old users.

[0028] Furthermore, in step S5, the method for obtaining the activity data of the new user in the model in real time is:

[0029] After a new user uses the model, real-time behavior data is generated, which can be recorded and represented by event logs: Y x (t)={y1(t), y2(t), y3(t),···y n (t),}, where Y x (t) is the activity log set of the new user x at time t, y1(t) is the first activity recorded at time t, n is the total number of activities, and the formula for feature extraction of the acquired real-time behavior data is expressed as: F x (t) = f[Y x (t)], where F x (t) is the feature vector of the new user x at time t, and f(·) is used to convert the original activity data Y x (t) is converted into a preprocessing function for feature vector.

[0030] Furthermore, in step S6, the method of continuously updating and optimizing the model through feedback data from new users is as follows:

[0031] After recording the acquired new user feature vector in the model, the parameter package of the new user in the model will be updated. During the update process, there will be a deviation between the real feedback feature vector data and the model prediction. The expression formula for measuring this deviation using the loss function is: in is the model's predicted score for new user x on the store IP address, u x，ip is the actual feedback of the new user x, s is the number of user feedback data, δ is the model parameter, and the optimization update formula for the model parameter δ is: where δ (i) is the parameter of the model at the i-th iteration, η is the learning rate, which controls the step size of each update, is the gradient of the loss function I with respect to the parameter δ.

[0032] Furthermore, in step S6, the method of continuously updating and optimizing the model through feedback data from new users is as follows:

[0033] As new user feedback data continues to accumulate, the model will continue to self-learn, optimize, and update. The formula for updating model parameters based on new feedback data is: (i+1) =δ (i) -η▽ δ I{δ (i) , F x (t)}, where F x (t) is the feature vector of new user feedback data.

[0034] Beneficial Effects

[0035] Compared with the known public technology, the technical solution provided by the present invention has the following advantages:

[0036] Beneficial effects:

[0037] The present invention obtains the data of new users and the activity data of new users on other platforms, analyzes these data and compares them with the historical data of old users. After obtaining similar features, the new users are classified, and the preferences of new users are predicted based on the preferences of old users for store food, and personalized store recommendations are made. The purchase and attention of new users to food on other platforms are used as the basis for associated recommendations, thereby improving the accuracy of recommended stores. At the same time, the model will obtain feedback data from new users in real time, and measure the deviation between the feedback data and the model predicted data. The model parameters are optimized and updated through the feedback data of new users, so that the model can better capture the user's preferences, gradually improve the accuracy of the present invention in predicting the preferences of new users, and further improve the accuracy of recommending stores to new users. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The present invention is a flowchart of a method for building a large model of food big data. DETAILED DESCRIPTION

[0039] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0040] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0041] The present invention is further described in detail with reference to the accompanying drawings:

[0042] Example:

[0043] like Figure 1 As shown, the present invention provides a method for building a large model of food big data, including the following steps: S1, obtaining old user historical data, store food data, new user basic data, social platform data and shopping platform data;

[0044] Furthermore, in step S1, the method for obtaining old user historical data, store food data, new user basic data, social platform data and shopping platform data is:

[0045] Access the user data module through the internal API to obtain user activity data. Store food data is obtained through the merchant's registration, and can also be obtained from user comments and feedback data. New user basic data is obtained during registration, and new users are guided to choose food preferences during registration. After the user authorizes the social platform data and shopping platform data, the user's social behavior, published content, shopping history, preference data and browsing data are obtained;

[0046] In this embodiment, when a new user registers on the model of the present invention, the basic information of the new user can be obtained, and the basic information includes but is not limited to age, gender, real-time geographic location and native location. At the same time, the new user is guided to choose the taste preference of food, and after the new user authorizes, the activity data of the new user on the design platform and the shopping platform are associated, and the data about diet in the activity data is marked.

[0047] S2. Process the basic data of the new user and the platform data related to the new user;

[0048] Furthermore, in step S2, the method for processing the basic data of the new user and the platform data related to the new user is:

[0049] Fill in the missing values ​​in the acquired new user basic data and standardize the user's numerical data. The filling formula is:

[0050] in is the estimated value of the missing value of user i in dimension j, is the global average of all user feature data, a i is the bias term of user i, representing the overall preference of the user, a j is the bias term of feature j, representing the overall preference of the feature, a k is the bias of other features k related to feature j, which is used to adjust the estimate m when calculating missing values ​​in collaborative filtering jk is the similarity weight between feature j and other features k, K is the set of all other features, y ik is the known value of feature k for user i, and the normalization formula is: Where Q ij is the standardized value of the i-th user on the j-th feature, y ij is the original data value of the i-th user on the j-th feature, m j is the mean of the jth feature, that is, the average value of all users on this feature: δ j is the standard deviation of the jth feature, that is, the degree of fluctuation of the feature: n is the total number of users. When processing multi-dimensional data, the feature matrix of the entire new user can be standardized. Let C be an n×t matrix, representing n users and t features, then the matrix formula is: Where B is the standardized matrix, m is a column vector containing the mean of each feature, with a size of 1×t, and δ is a column vector containing the standard deviation of each feature, with a size of 1×t, which standardizes the feature data of all users;

[0051] In this embodiment, the data of new users is preprocessed to fill in missing values ​​in the new user data, and the feature data of new users is standardized to improve the accuracy and rationality of the new user data, thereby providing rich new user portraits and activity data for the present invention.

[0052] S3. Analyze the activity data of new users on social platforms and shopping platforms;

[0053] Furthermore, in step S3, the method for analyzing the activity data of the new user on the social platform and the shopping platform is:

[0054] Correlate and mine the data of new users on social platforms and shopping platforms, clean the data after acquisition and form an event log table, analyze the click path, dwell time and conversion rate of new users' activity data, compare and identify the data of new users with the data of old users, including but not limited to gender, age, geographic location, and purchased food, so as to identify similar types of user groups and classify them;

[0055] Furthermore, in step S3, the method for analyzing the activity data of the new user on the social platform and the shopping platform is:

[0056] The process of analyzing the click path, dwell time, and conversion rate in comparative identification is as follows: i =[q i1 ,q i2 ,q i3 ,···q in ], where P i is the click sequence of the i-th user, q in is the nth click behavior of the user, including but not limited to product pages, activity pages and food videos; the dwell time analysis formula is: where S i The total stay time of user i, S in is the user's stay time on page n, and the conversion rate analysis formula is: Where Z i is the ratio of user i from click to actual purchase, k is the number of conversion events, and N is the total number of clicks. By analyzing the dwell time of new users on different pages and the purchase conversion rate, the preferences of new users can be identified;

[0057] Furthermore, in step S3, the method for analyzing the activity data of the new user on the social platform and the shopping platform is:

[0058] The similarity calculation formula is: Where x is a new user, y is an old user, and u x，n and u y，n is the value of feature n of user x and user y. The value range of cosine similarity is [-1, 1]. The closer the value is to 1, the more similar they are. The classification formula based on similarity is: where H is the number of clusters, Z h is the user set of the hth cluster, v i is the feature vector of the i-th user, κ h is the center point of the hth cluster. By analyzing the new user activity data, the new user group can be identified and classified;

[0059] In this embodiment, analyzing the activity data of new users is an important part of the present invention in constructing new user portraits. The personal information, preference information, current location and hometown location in the new user's basic data can provide a basis for constructing the new user's initial portrait, and classification can be performed based on the new user's basic data to make recommendations for different groups of new users. For new users who provide food taste preferences when registering, corresponding food stores can be matched for analysis and recommendation. By analyzing the new user's hometown location, analysis can be performed based on the traditional food and tastes in the hometown location and combined with the current location. Through these methods, a rich and specific initial portrait of the new user can be constructed.

[0060] S4. Collaboratively filter the analyzed data with the historical data of old users and the food data of the store;

[0061] Furthermore, in step S4, the method of collaboratively filtering the analyzed data with the old user historical data and the store food data is as follows:

[0062] After analyzing new users, we can identify the similarities between new users and old users. Based on the similarities between new users and old users, and based on the old users' preferences for store food, we can predict the food stores preferred by new users. The prediction formula is: in is the score of store i predicted for new user x, x n is the historical average rating of new user x, Z x is the set of old users similar to the new user x, sim(x, y) is the similarity between the new user x and the old user y, is the actual rating of old user y on store i, l y is the mean rating of old user y. By considering the ratings and comments of old users on the store, we can predict the ratings and comments of new users who are similar to old users.

[0063] In this embodiment, after analyzing the activity data of the new user, the feature data of the new user is obtained, and then the feature data is analyzed and compared with the historical data of the old user. It is very likely that similar old users can be found, and then combined with the old users' historical ratings and comments on different food stores, the new user's ratings for stores that have not been visited are predicted, and finally a recommendation result is generated for the new user, thereby providing the best recommendation result to the new user.

[0064] S5, obtain the activity data of new users in the model in real time;

[0065] Furthermore, in step S5, the method for obtaining the activity data of the new user in the model in real time is:

[0066] After a new user uses the model, real-time behavior data is generated, which can be recorded and represented by event logs:x (t)={y1(t), y2(t), y3(t),···y n (t),}, where Y x (t) is the activity log set of the new user x at time t, y1(t) is the first activity recorded at time t, n is the total number of activities, and the formula for feature extraction of the acquired real-time behavior data is expressed as: F x (t) = f[Y x (t)], where F x (t) is the feature vector of the new user x at time t, and f(·) is used to convert the original activity data Y x (t) Preprocessing function converted into feature vector;

[0067] In this embodiment, after a new user registers on the model of the present invention, the model will obtain the new user's activity data in real time, including but not limited to clicks, dwell time, comments, ratings, favorites, follows and page views. After obtaining these real-time activity data, the user's preference model is updated, and the feature vector of the new user's activity data is extracted to provide a data set for subsequent model updates and optimizations.

[0068] S6. Continuously update and optimize the model through feedback data from new users;

[0069] Furthermore, in step S6, the method of continuously updating and optimizing the model through feedback data from new users is as follows:

[0070] After recording the acquired new user feature vector in the model, the parameter package of the new user in the model will be updated. During the update process, there will be a deviation between the real feedback feature vector data and the model prediction. The expression formula for measuring this deviation using the loss function is: in is the model's predicted score for new user x on the store IP address, u x，ip is the actual feedback of the new user x, s is the number of user feedback data, δ is the model parameter, and the optimization update formula for the model parameter δ is: where δ (i) is the parameter of the model at the i-th iteration, η is the learning rate, which controls the step size of each update, ▽ δ I{δ (i)} is the gradient of the loss function I with respect to the parameter δ;

[0071] Furthermore, in step S6, the method of continuously updating and optimizing the model through feedback data from new users is as follows:

[0072] As new user feedback data continues to accumulate, the model will continue to self-learn, optimize, and update. The formula for updating model parameters based on new feedback data is:(i+1) =δ (i) -η▽ δ I{δ (i) , F x (t)}, where F x (t) is the feature vector of new user feedback data

[0073] In this embodiment, there will be a deviation between the real-time behavior of new users and the model prediction. By measuring the deviation, adjustments and optimizations can be made based on the user's real-time feedback data, gradually improving the ability to predict user preferences, thereby further improving the accuracy of the large model in recommending food stores to new users after modeling.

[0074] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for building a large model of food big data, characterized in that: The following steps are involved: S1. Obtain old user historical data, store food data, new user basic data, social platform data and shopping platform data; S2. Process the basic data of the new user and the platform data related to the new user; S3. Analyze the activity data of new users on social platforms and shopping platforms; S4. Collaboratively filter the analyzed data with the historical data of old users and the food data of the store; S5, obtain the activity data of new users in the model in real time; S6. Continuously update and optimize the model through feedback data from new users.

2. The method for building a large model of food big data according to claim 1, characterized in that: In step S1, the method for obtaining old user historical data, store food data, new user basic data, social platform data and shopping platform data is: Access the user data module through the internal API to obtain user activity data. Store food data is obtained through the merchant's registration, and can also be obtained from user comments and feedback data. New user basic data is obtained during registration, and new users are guided to choose food preferences during registration. After the user authorizes the social platform data and shopping platform data, the user's information including but not limited to social behavior, published content, shopping history, preference data and browsing data is obtained.

3. The method for building a large model of food big data according to claim 2, characterized in that: In step S2, the method for processing the basic data of the new user and the platform data related to the new user is: Fill in the missing values ​​in the acquired new user basic data and standardize the user's numerical data. The filling formula is: in is the estimated value of the missing value of user i in dimension j, is the global average of all user feature data, a i is the bias term of user i, representing the overall preference of the user, a j is the bias term of feature j, representing the overall preference of the feature, a k is the bias of other features k related to feature j, which is used to adjust the estimate m when calculating missing values ​​in collaborative filtering jk is the similarity weight between feature j and other features k, K is the set of all other features, y ik is the known value of feature k for user i, and the normalization formula is: Where Q ij is the standardized value of the i-th user on the j-th feature, y ij is the original data value of the i-th user on the j-th feature, m j is the mean of the jth feature, that is, the average value of all users on this feature: δ j is the standard deviation of the jth feature, that is, the degree of fluctuation of the feature: n is the total number of users. When processing multi-dimensional data, the feature matrix of the entire new user can be standardized. Let C be an n×t matrix, representing n users and t features, then the matrix formula is: Where B is the standardized matrix, m is a column vector containing the mean of each feature with a size of 1×t, and δ is a column vector containing the standard deviation of each feature with a size of 1×t, which standardizes the feature data of all users.

4. The method for building a large model of food big data according to claim 3 is characterized in that: In step S3, the method for analyzing the activity data of the new user on the social platform and the shopping platform is: The data of new users on social platforms and shopping platforms are associated and mined. After obtaining the data, the data is cleaned and formed into an event log table. The click path, dwell time and conversion rate of the new users' activity data are analyzed. The data of new users are compared and identified with the data of old users. The comparison and identification content includes but is not limited to gender, age, geographic location, and purchased food, so as to identify similar types of user groups and classify them.

5. The method for building a large model of food big data according to claim 4, characterized in that: In step S3, the method for analyzing the activity data of the new user on the social platform and the shopping platform is: The process of analyzing the click path, dwell time, and conversion rate in comparative identification is as follows: i =[q i1 ,q i2 ,q i3 ,···q in ], where P i is the click sequence of the i-th user, q in is the nth click behavior of the user, including but not limited to product pages, activity pages and food videos; the dwell time analysis formula is: where S i The total stay time of user i, S in is the user's stay time on page n, and the conversion rate analysis formula is: Where Z i is the ratio of user i from click to actual purchase, k is the number of conversion events, and N is the total number of clicks. By analyzing the dwell time of new users on different pages and the purchase conversion rate, the preferences of new users can be identified.

6. The method for building a large model of food big data according to claim 5, characterized in that: In step S3, the method for analyzing the activity data of the new user on the social platform and the shopping platform is: The similarity calculation formula is: Where x is a new user, y is an old user, and u x，n and u y，n is the value of feature n of user x and user y. The value range of cosine similarity is [-1, 1]. The closer the value is to 1, the more similar they are. The classification formula based on similarity is: where H is the number of clusters, Z h is the user set of the hth cluster, v i is the feature vector of the i-th user, κ h It is the center point of the hth cluster. By analyzing the activity data of new users, the new user groups can be identified and classified.

7. The method for building a large model of food big data according to claim 6, characterized in that: In step S4, the method for collaboratively filtering the analyzed data with the old user historical data and the store food data is as follows: After analyzing new users, we can identify the similarities between new users and old users. Based on the similarities between new users and old users, and based on the old users' preferences for store food, we can predict the food stores preferred by new users. The prediction formula is: in is the score of store i predicted for new user x, x n is the historical average rating of new user x, Z x is the set of old users similar to the new user x, sim(x, y) is the similarity between the new user x and the old user y, is the actual rating of old user y on store i, l y is the mean rating of old users y. By considering the ratings and comments of old users on the store, we can predict the ratings and comments of new users who are similar to old users.

8. The method for building a large model of food big data according to claim 7, characterized in that: In step S5, the method for obtaining the activity data of the new user in the model in real time is: After a new user uses the model, real-time behavior data is generated, which can be recorded and represented by event logs: x (t)={y1(t), y2(t), y3(t),···y n (t),}, where Y x (t) is the activity log set of the new user x at time t, y1(t) is the first activity recorded at time t, n is the total number of activities, and the formula for feature extraction of the acquired real-time behavior data is expressed as: F x (t) = f[Y x (t)], where F x (t) is the feature vector of the new user x at time t, and f(·) is used to convert the original activity data Y x (t) is converted into a preprocessing function for feature vector.

9. The method for building a large model of food big data according to claim 8, characterized in that: In step S6, the method of continuously updating and optimizing the model through feedback data from new users is: After recording the acquired new user feature vector in the model, the parameter package of the new user in the model will be updated. During the update process, there will be a deviation between the real feedback feature vector data and the model prediction. The expression formula for measuring this deviation using the loss function is: in is the model's predicted score for new user x on the store IP address, u x，ip is the actual feedback of the new user x, s is the number of user feedback data, δ is the model parameter, and the optimization update formula for the model parameter δ is: where δ (i) is the parameter of the model at the i-th iteration, η is the learning rate, which controls the step size of each update, is the gradient of the loss function I with respect to the parameter δ.

10. The method for building a large model of food big data according to claim 8, characterized in that: In step S6, the method of continuously updating and optimizing the model through feedback data from new users is: As new user feedback data continues to accumulate, the model will continue to self-learn, optimize, and update. The formula for updating model parameters based on new feedback data is: where F x (t) is the feature vector of new user feedback data.

Images