Personalized customization pushing system and method for user portrait technology

Abstract

The invention discloses a personalized customization pushing system and method of a user portrait technology, and relates to the technical field of personalized pushing, a terminal obtains a user portrait through authorization of a target user and uploads non-privacy information in the target user portrait to a server, and the server generates a universal recommendation result of the target user; the server sends the universality recommendation result to the terminal, the terminal obtains a particularity recommendation result based on the universality recommendation result and the target user portrait, and the terminal obtains a personalized customization pushing method according to the universality recommendation result and the particularity recommendation result; transmission of the user portrait containing privacy information is not involved, and the risk of user portrait information leakage in the transmission process is reduced; transmission of personalized customization push results is not involved, and the risk that personalized customization push content exposes privacy information of a user is reduced; and the privacy information of the user is reserved in the terminal equipment, so that the security of the user information is improved.

Description

Technical Field

[0001] The present invention relates to the field of personalized push technology, and in particular to a personalized customized push system and method based on user portrait technology. Background Art

[0002] User portrait is a commonly used concept in data analysis, marketing, product design and other fields. It collects and analyzes various user characteristics, behaviors, preferences and other information to construct a virtual, representative user model, which is used to better understand user needs, optimize products and services, and formulate marketing strategies. By constructing user portraits, companies can better understand user needs, preferences and behaviors, thereby achieving accurate advertising and personalized recommendations. User portraits rely on a large amount of user data. During the data transmission process, if this data is improperly collected, stored or shared, it may lead to user privacy leakage. In addition, the personalized recommendation content provided to users based on user portraits also has security and privacy issues. Therefore, how to protect user privacy while applying user portrait technology for personalized customized push has become an urgent problem that needs to be solved. Summary of the Invention

[0003] The purpose of the present invention is to provide a personalized customized push system and method based on user portrait technology to solve the problems raised in the prior art.

[0004] To achieve the above-mentioned objectives, the present invention provides the following technical solutions: a personalized customized push system based on user portrait technology, comprising a terminal and a server; the terminal and the server are interconnected, obtain a target user portrait through authorization, and transmit the non-private information in the target user portrait to the server; the server determines the target user's group information based on the non-private information in the target user portrait, determines the target user's universal recommendation results based on the target user's group information, and sends the result to the terminal; the terminal obtains the target user's specific recommendation results based on the target user's universal recommendation results and the target user portrait, and obtains personalized customized push based on the target user's specific recommendation results and universal recommendation results.

[0005] Specifically, the terminal also includes an information collection module, a first communication module, a first data storage module, a particularity recommendation module, a personalized customization push module and a deviation analysis module; the information collection module is used to obtain a target user portrait; the first communication module is used to establish a connection with a server; the first data storage module is used to store the user portrait of the target user; the particularity recommendation module obtains a particularity recommendation result for the target user based on the universal recommendation result of the target user and the target user portrait; the personalized customization push module obtains a personalized customization push based on the particularity recommendation result and the universal recommendation result of the target user; the deviation analysis module is used to determine the deviation between the personalized customization push and the universal recommendation result. The server also includes a second communication module, a second data storage module, a group analysis module, a universal recommendation module, a group update module and a model generation module; the second communication module is used to realize data interaction between the server and the terminal; the second data storage module is used to store historical user portraits; the group analysis module determines the group information of the target user based on the necessary characteristics of the target user; the universal recommendation module obtains the universal recommendation results of the target user based on the group information of the target user; the group update module is used to determine the necessary characteristics of the target user, and update the necessary characteristics of the target user according to the deviation between the personalized customized push and the universal recommendation results; the model generation module is used to train the first recommendation model and the second recommendation model; the first recommendation model obtains the recommendation results based on the user portrait, and the second recommendation model corrects the recommendation results of the first recommendation model based on the recommendation results of the first recommendation model through additional privacy information.

[0006] Specifically, the server trains a second recommendation model based on the first recommendation model, takes the historical user portrait containing privacy information as the advanced portrait, and the historical user portrait not containing privacy information as the basic portrait, and recommends the basic portrait and the advanced portrait through the first recommendation model to obtain the basic recommendation result and the advanced recommendation result. According to the deviation between the basic recommendation result and the advanced recommendation result, the influence of the privacy information on the recommendation is obtained; the server takes the basic recommendation result and the influence of the privacy information on the recommendation as input, and the advanced recommendation result as output for training to obtain the second recommendation model; the server sends the universal recommendation result and the second recommendation model to the terminal, and the terminal obtains the specific recommendation result based on the privacy information in the locally stored user portrait using the second recommendation model and the universal recommendation result; and obtains personalized customized push based on the universal recommendation result and the specific recommendation result.

[0007] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a personalized customized push method based on user portrait technology, comprising the following steps: The terminal obtains the user portrait through authorization from the target user and uploads the non-private information in the target user portrait to the server. The server then generates universal recommendation results for the target user. The server sends the universal recommendation results to the terminal, and the terminal obtains the specific recommendation results based on the universal recommendation results and the target user portrait. The terminal obtains a personalized push method based on the universal recommendation results and the specific recommendation results.

[0008] Specifically, the server generating the user's universal recommendation result further includes the following steps: The terminal obtains the user portrait through authorization, and the server obtains the non-private information of the target user from the terminal, and determines the necessary features from the non-private information of the target user based on the first recommendation model and the historical user portrait in the server; Let x i represents the i-th feature in the target user's non-private information. The server selects the features whose differences only exist in feature x from the historical user profiles. i The historical user portraits are divided into the first category of user portraits and the second category of user portraits, where the first category of user portraits includes the feature x i , the second type of user portrait does not include feature x i The server uses the first recommendation model to recommend the first category user profile and the second category user profile respectively, obtains the first category recommendation results and the second category recommendation results, calculates the root mean square error between the first category recommendation results and the second category recommendation results, and if the root mean square error is greater than the error threshold, then the i-th feature in the target user's non-private information is regarded as a necessary feature; if the error is not greater than the error threshold, then the i-th feature in the target user's non-private information is regarded as a required feature; Historical user profiles include both those with and without private information. The server can expand the private information in historical user profiles based on the historical user profiles without private information to generate user profiles with private information. Alternatively, the user can upload the user profile from the terminal to the server with the historical user's authorization. With the user's knowledge and consent, the user can upload the user profile with private information to the server to obtain better personalized push results. Necessary features have a greater impact on recommendation results, while optional features have a lower impact on recommendation results. Necessary features are used to determine user group information. The first recommendation model obtains recommendation results based on user portraits. The second recommendation model uses additional privacy information to modify the recommendation results of the first recommendation model. The first recommendation model can also include privacy information, but the target user does not need to upload their own privacy information to the server. This does not involve the transmission of the target user's privacy information, reducing the risk of data leakage. The server recommends stored historical user profiles using a first recommendation model to obtain recommendation results. The first recommendation model inputs the features in the user profiles and outputs recommendation scores for all services to be recommended. The recommendation scores of all services to be recommended are used as input vectors and clustered using the DBSCAN algorithm to obtain clusters. Each cluster contains several historical user profiles. All necessary features in the target user's non-private information are obtained, and a cluster is determined that includes all necessary features in the target user's non-private information. As long as there is a historical user profile in the cluster that includes all necessary features in the target user's non-private information, the cluster also includes all necessary features in the target user's non-private information. Based on the determined cluster, the group information that the target user meets is obtained.

[0009] Specifically, the server generating the user's universal recommendation result further includes the following steps: The server recommends each group that the target user matches based on the target user’s group information, and obtains the group recommendation result T j , where T j represents the recommendation result of the jth group; sum the recommendation results of all groups that the target user meets, and get the universal recommendation result T of the target user, T=∑(W j ×T j ), where W j The weight of the jth group that the target user meets is calculated using the following formula: , where n j N represents the number of historical user profiles in the jth group that contain all necessary features of the target user’s non-private information. j Represents the number of historical user portraits in the j-th group.

[0010] Specifically, the method for obtaining personalized push notifications further includes the following steps: The server trains a second recommendation model based on the first recommendation model, using historical user portraits containing privacy information as advanced portraits and historical user portraits that do not contain privacy information as basic portraits. The first recommendation model is used to recommend the basic and advanced portraits to obtain basic and advanced recommendation results. Based on the deviation between the basic and advanced recommendation results, the impact of privacy information on the recommendation is obtained. The server uses the basic recommendation results and the impact of privacy information on the recommendation as input and the advanced recommendation results as output for training to obtain the second recommendation model.

[0011] The impact of privacy information on recommendations can be determined by controlling variables, that is, controlling that the advanced profile contains all the non-privacy features of the basic profile, the difference only lies in the privacy features. The impact of privacy information on recommendations is obtained based on the recommendation results after the privacy features are included.

[0012] Specifically, the method for obtaining personalized push notifications further includes the following steps: The server sends the universal recommendation results and the second recommendation model to the terminal. The terminal uses the second recommendation model and the universal recommendation results to obtain the specific recommendation results based on the privacy information in the locally stored user portrait. The terminal obtains the personalized push G based on the universal recommendation results and the specific recommendation results, where G=U×T+V×E, where U and V are the weights of the universal recommendation results and the specific recommendation results, and E is the specific recommendation result for the target user.

[0013] Specifically, the method for obtaining personalized push notifications further includes the following steps: Based on the target user's group information, features included in the group information but not included in the target user's portrait and the target user's required features are obtained as identification features, and the impact of the identification features on the recommendation is determined; the matching degree between the impact of each identification feature on the recommendation and the deviation is calculated to obtain the identification feature with the highest matching degree. If the matching degree is greater than a threshold, the identification feature with the highest matching degree is added to the target user's main features, and the target user's group information is updated based on the target user's main features; the server updates the target user's universal recommendation results according to the updated target user's group information and sends them to the terminal.

[0014] The impact of identification features on recommendations can be determined by referring to the impact of privacy information on recommendations.

[0015] Compared with the existing technology, the beneficial effects of the present invention are: it does not involve the transmission of user portraits containing privacy information, reducing the risk of leakage of user portrait information during transmission; it does not involve the transmission of personalized push results, reducing the risk of personalized push content exposing user privacy information; the user's privacy information is retained in the terminal device, improving the security of user information. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the structure of the personalized customized push system using the user portrait technology of the present invention; Figure 2 This is a flowchart of the personalized customized push method based on the user portrait technology of the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0018] Example: Figure 1 As shown, the present invention provides a technical solution, a personalized push system using user portrait technology, including a terminal and a server; the terminal and the server are interconnected, obtain the target user portrait through authorization, and transmit the non-private information in the target user portrait to the server; the server determines the target user's group information based on the non-private information in the target user portrait, determines the target user's universal recommendation results based on the target user's group information, and sends them to the terminal; the terminal obtains the target user's specific recommendation results based on the target user's universal recommendation results and the target user portrait, and obtains personalized push based on the target user's specific recommendation results and the universal recommendation results. Terminals include but are not limited to mobile phones, tablets, computers, etc., and user tags are obtained during user use.

[0019] Specifically, the terminal also includes an information collection module, a first communication module, a first data storage module, a particularity recommendation module, a personalized customization push module and a deviation analysis module; the information collection module is used to obtain a target user portrait; the first communication module is used to establish a connection with a server; the first data storage module is used to store the user portrait of the target user; the particularity recommendation module obtains a particularity recommendation result for the target user based on the universal recommendation result of the target user and the target user portrait; the personalized customization push module obtains a personalized customization push based on the particularity recommendation result and the universal recommendation result of the target user; the deviation analysis module is used to determine the deviation between the personalized customization push and the universal recommendation result. The server also includes a second communication module, a second data storage module, a group analysis module, a universal recommendation module, a group update module and a model generation module; the second communication module is used to realize data interaction between the server and the terminal; the second data storage module is used to store historical user portraits; the group analysis module determines the group information of the target user based on the necessary characteristics of the target user; the universal recommendation module obtains the universal recommendation results of the target user based on the group information of the target user; the group update module is used to determine the necessary characteristics of the target user, and update the necessary characteristics of the target user according to the deviation between the personalized customized push and the universal recommendation results; the model generation module is used to train the first recommendation model and the second recommendation model; the first recommendation model obtains the recommendation results based on the user portrait, and the second recommendation model corrects the recommendation results of the first recommendation model based on the recommendation results of the first recommendation model through additional privacy information.

[0020] Specifically, the server trains a second recommendation model based on the first recommendation model, takes the historical user portrait containing privacy information as the advanced portrait, and the historical user portrait not containing privacy information as the basic portrait, and recommends the basic portrait and the advanced portrait through the first recommendation model to obtain the basic recommendation result and the advanced recommendation result. According to the deviation between the basic recommendation result and the advanced recommendation result, the influence of the privacy information on the recommendation is obtained; the server takes the basic recommendation result and the influence of the privacy information on the recommendation as input, and the advanced recommendation result as output for training to obtain the second recommendation model; the server sends the universal recommendation result and the second recommendation model to the terminal, and the terminal obtains the specific recommendation result based on the privacy information in the locally stored user portrait using the second recommendation model and the universal recommendation result; and obtains personalized customized push based on the universal recommendation result and the specific recommendation result.

[0021] Example: Figure 2 As shown, the present invention provides a technical solution, a personalized customized push method of user portrait technology, comprising the following steps: The terminal obtains the user portrait through the target user's authorization and uploads the non-private information in the target user portrait to the server. The server then generates a general recommendation result for the target user: Let x i represents the i-th feature in the target user's non-private information. The server selects the features whose differences only exist in feature x from the historical user profiles. i The historical user portraits are divided into the first category of user portraits and the second category of user portraits, where the first category of user portraits includes the feature x i , the second type of user portrait does not include feature x i The server uses the first recommendation model to recommend the first category user profile and the second category user profile respectively, obtains the first category recommendation results and calculates the root mean square error between the first category recommendation results and the second category recommendation results. If the root mean square error is greater than the error threshold, the i-th feature in the target user's non-private information is regarded as a necessary feature; if the root mean square error is not greater than the error threshold, the i-th feature in the target user's non-private information is regarded as an optional feature. The output results of the first recommendation model and the second recommendation model are recommendation scores with recommended businesses. For example, if an enterprise involves a total of 5 businesses from A1 to A5, the first recommendation model and the second recommendation model will obtain the recommendation scores of these 5 businesses, obtain the weight of each business based on the recommendation scores, and display the enterprise's businesses to users based on the weights.

[0022] The server recommends stored historical user profiles using a first recommendation model to obtain recommendation results. The first recommendation model inputs the features in the user profiles and outputs recommendation scores for all services to be recommended. The recommendation scores of all services to be recommended are used as input vectors and clustered using the DBSCAN algorithm to obtain clusters. Each cluster contains several historical user profiles. The specific steps of clustering are as follows: Obtain the recommendation results of the first recommendation model for the historical user profile, that is, the recommendation scores of the five businesses. Use the recommendation scores as input vectors for DBSCAN clustering. First, set the neighborhood radius and minimum number of points. The neighborhood radius and minimum number of points can be tested with different combinations. Step 1: Select an input vector as the starting point; Step 2: Taking the selected input vector as the center, search for other input vectors within the neighborhood radius; Step 3: If the number of other input vectors found within the neighborhood radius is not less than the minimum number of points, the selected input vector is marked as a core point, and a cluster is formed with this core point; if the number of other input vectors found within the neighborhood radius is less than the minimum number of points, the input vector is marked as a noise point; Step 4: For the core point, all input vectors within the neighborhood radius centered on the core point are added to the cluster formed by the core point; for other input vectors added to the core point cluster, determine whether they are core points according to step 3. If they are core points, continue to expand the cluster; if they are not core points, do not expand; Repeat steps 1 to 4 until all input vectors have been visited.

[0023] All necessary features in the target user's non-private information are obtained, and a cluster is determined that includes all necessary features in the target user's non-private information. As long as there is a historical user profile in the cluster that includes all necessary features in the target user's non-private information, the cluster also includes all necessary features in the target user's non-private information. Based on the determined cluster, the group information that the target user meets is obtained.

[0024] The server recommends each group that the target user matches based on the target user’s group information, and obtains the group recommendation result T j , where T j represents the recommendation result of the jth group; sum the recommendation results of all groups that the target user meets, and get the universal recommendation result T of the target user, T=∑(W j ×T j ), where W j The weight of the jth group that the target user meets is calculated using the following formula: , where nj N represents the number of historical user profiles in the jth group that contain all necessary features of the target user’s non-private information. j Represents the number of historical user portraits in the j-th group.

[0025] The server sends the universal recommendation results to the terminal, and the terminal obtains the specific recommendation results based on the universal recommendation results and the target user portrait. The terminal obtains a personalized push method based on the universal recommendation results and the specific recommendation results.

[0026] The server trains a second recommendation model based on the first recommendation model, using historical user portraits containing privacy information as advanced portraits and historical user portraits that do not contain privacy information as basic portraits. The first recommendation model is used to recommend the basic and advanced portraits to obtain basic and advanced recommendation results. Based on the deviation between the basic and advanced recommendation results, the impact of privacy information on the recommendation is obtained. The server uses the basic recommendation results and the impact of privacy information on the recommendation as input and the advanced recommendation results as output for training to obtain the second recommendation model.

[0027] The server sends the universal recommendation results and the second recommendation model to the terminal. The terminal uses the second recommendation model and the universal recommendation results to obtain the specific recommendation results based on the privacy information in the locally stored user portrait. The terminal obtains the personalized push G based on the universal recommendation results and the specific recommendation results, where G=U×T+V×E, where U and V are the weights of the universal recommendation results and the specific recommendation results, and E is the specific recommendation result for the target user.

[0028] The terminal obtains the deviation based on the personalized customized push G and the universal recommendation result T of the target user and feeds the deviation back to the server; Based on the target user's group information, features included in the group information but not included in the target user's portrait and the target user's required features are obtained as identification features, and the impact of the identification features on the recommendation is determined; the matching degree between the impact of each identification feature on the recommendation and the deviation is calculated to obtain the identification feature with the highest matching degree. If the matching degree is greater than a threshold, the identification feature with the highest matching degree is added to the target user's main features, and the target user's group information is updated based on the target user's main features; the server updates the target user's universal recommendation results according to the updated target user's group information and sends them to the terminal.

[0029] The deviation between the personalized push G and the universal recommendation result T is the deviation between the recommendation scores of the five businesses, which can be reflected by the root mean square error; according to the impact of each identification feature on the recommendation, for example, when the identification feature B1 is used, the recommendation score of business A1 will increase. The user's necessary features do not yet include the identification feature B1. When the recommendation score of business A1 increases, the deviation between the personalized push G and the universal recommendation result T decreases. Then, the contribution of the identification feature B1 to reducing the deviation can be obtained. Based on the contribution, the matching degree between B1 and the deviation is obtained. When the matching degree is high enough, B1 is added to the necessary features.

[0030] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A personalized customized push method based on user portrait technology, characterized in that: The following steps are involved: The terminal obtains the user portrait through authorization of the target user, uploads the non-privacy information in the target user portrait to the server, and the server generates a universal recommendation result for the target user; The server sends the universal recommendation results to the terminal, and the terminal obtains the specific recommendation results based on the universal recommendation results and the target user portrait. The terminal obtains a personalized push method based on the universal recommendation results and the specific recommendation results.

2. The personalized customized push method based on user portrait technology according to claim 1 is characterized in that: The server generates a universal recommendation result for the target user and further comprises the following steps: The terminal obtains the target user portrait through authorization, the server obtains the target user's non-private information from the terminal, and determines necessary features from the target user's non-private information based on the first recommendation model and the historical user portrait in the server; Let x i represents the i-th feature in the target user’s non-private information. The server selects the features whose differences only exist in feature x from the historical user profiles. i The historical user portraits are divided into the first category of user portraits and the second category of user portraits, where the first category of user portraits includes the feature x i , the second type of user portrait does not include feature x i The server uses the first recommendation model to recommend the first type of user portrait and the second type of user portrait respectively, obtains the first type of recommendation results and the second type of recommendation results, calculates the root mean square error between the first type of recommendation results and the second type of recommendation results, and if the root mean square error is greater than the error threshold, the i-th feature in the target user's non-private information is taken as a necessary feature; if the error is not greater than the error threshold, the i-th feature in the target user's non-private information is taken as a required feature; The server recommends the stored historical user portraits through the first recommendation model to obtain a recommendation result; the input of the first recommendation model is the features in the user portrait, and the output is the recommendation score of all the services to be recommended; the recommendation scores of all the services to be recommended are used as input vectors, and clustering is performed through the DBSCAN algorithm to obtain clusters; a number of historical user portraits in each cluster; All necessary features in the non-private information of the target user are obtained, and a cluster including all necessary features in the non-private information of the target user is determined. As long as there is a historical user profile in the cluster including all necessary features in the non-private information of the target user, the cluster also includes all necessary features in the non-private information of the target user. According to the determined cluster, the group information that the target user meets is obtained.

3. The personalized customized push method based on user portrait technology according to claim 2 is characterized in that: The server generates a universal recommendation result for the user, and further comprises the following steps: The server recommends each group that the target user matches based on the target user’s group information, and obtains the group recommendation result T j , where T j represents the recommendation result of the jth group; sum the recommendation results of all groups that the target user meets, and get the universal recommendation result T of the target user, T=∑(W j ×T j ), where W j The weight of the jth group that the target user meets is calculated using the following formula: , where n j represents the number of historical user profiles in the jth group that contain all necessary features of the target user’s non-private information, N j Represents the number of historical user portraits in the jth group.

4. The personalized customized push method based on user portrait technology according to claim 3 is characterized in that: The method for obtaining personalized push notifications also includes the following steps: The server trains a second recommendation model based on the first recommendation model, takes the historical user portrait containing privacy information as the advanced portrait, and the historical user portrait not containing privacy information as the basic portrait, and recommends the basic portrait and the advanced portrait through the first recommendation model to obtain basic recommendation results and advanced recommendation results. According to the deviation between the basic recommendation results and the advanced recommendation results, the influence of privacy information on the recommendation is obtained; the server takes the basic recommendation results and the influence of privacy information on the recommendation as input, and trains the advanced recommendation results as output to obtain the second recommendation model.

5. The personalized customized push method based on user portrait technology according to claim 4 is characterized in that: The method for obtaining personalized push notifications also includes the following steps: The server sends the universal recommendation result and the second recommendation model to the terminal. The terminal obtains the specific recommendation result based on the privacy information in the user portrait stored locally by using the second recommendation model and the universal recommendation result. The personalized push G is obtained according to the universal recommendation result and the specific recommendation result, G=U×T+V×E, where U and V are the weights of the universal recommendation result and the specific recommendation result, and E is the specific recommendation result for the target user.

6. The personalized customized push method based on user portrait technology according to claim 5 is characterized in that: The method for obtaining personalized push notifications also includes the following steps: The terminal obtains the deviation based on the personalized customized push G of the target user and the universal recommendation result T, and feeds the deviation back to the server; According to the group information of the target user, the features included in the group information but not included in the target user portrait and the required features of the target user are obtained as identification features, and the influence of the identification features on the recommendation is determined; the matching degree between the influence and deviation of each identification feature on the recommendation is calculated to obtain the identification feature with the highest matching degree. If the matching degree is greater than a threshold, the identification feature with the highest matching degree is added to the main feature of the target user, and the group information of the target user is updated according to the main feature of the target user; the server updates the universal recommendation result of the target user according to the updated group information of the target user and sends it to the terminal.

7. The personalized customized push system based on user portrait technology is characterized by: It includes a terminal and a server; the terminal is connected to the server, obtains a target user portrait through authorization, and transmits non-private information in the target user portrait to the server; the server determines the group information of the target user based on the non-private information in the target user portrait, determines the universal recommendation result of the target user based on the group information of the target user, and sends it to the terminal; The terminal obtains a specific recommendation result for the target user based on the general recommendation result for the target user and the target user portrait, and obtains a personalized customized push based on the specific recommendation result and the general recommendation result for the target user.

8. The personalized customized push system based on user portrait technology according to claim 7 is characterized in that: The terminal further includes an information collection module, a first communication module, a first data storage module, a particularity recommendation module, a personalized customization push module and a deviation analysis module; the information collection module is used to obtain a target user portrait; the first communication module is used to establish a connection with a server; the first data storage module is used to store a user portrait of the target user; The specific recommendation module obtains the specific recommendation result of the target user based on the universal recommendation result of the target user and the target user portrait; The personalized push module obtains personalized push based on the specific recommendation results and universal recommendation results of the target user; The deviation analysis module is used to determine the deviation between the personalized customized push and the universal recommendation results.

9. The personalized customized push system based on user portrait technology according to claim 8, characterized in that: The server further includes a second communication module, a second data storage module, a group analysis module, a universal recommendation module, a group update module and a model generation module; the second communication module is used to realize data interaction between the server and the terminal; the second data storage module is used to store historical user portraits; the group analysis module determines the group information of the target user based on the necessary characteristics of the target user; The universal recommendation module obtains universal recommendation results for target users based on group information of target users; The group update module is used to determine the necessary characteristics of the target user and update the necessary characteristics of the target user according to the deviation between the personalized customized push and the universal recommendation results; The model generation module is used to train a first recommendation model and a second recommendation model; the first recommendation model obtains a recommendation result based on a user portrait, and the second recommendation model corrects the recommendation result of the first recommendation model based on the recommendation result of the first recommendation model by using additional privacy information.

10. The personalized customized push system based on user portrait technology according to claim 9, characterized in that: The server trains a second recommendation model based on the first recommendation model, uses the historical user portrait containing privacy information as the advanced portrait, and the historical user portrait not containing privacy information as the basic portrait, recommends the basic portrait and the advanced portrait through the first recommendation model, obtains the basic recommendation result and the advanced recommendation result, and obtains the influence of privacy information on the recommendation according to the deviation between the basic recommendation result and the advanced recommendation result; the server uses the basic recommendation result and the influence of privacy information on the recommendation as input, and uses the advanced recommendation result as output for training, to obtain the second recommendation model; The server sends the universal recommendation result and the second recommendation model to the terminal, and the terminal obtains the specific recommendation result by using the second recommendation model and the universal recommendation result based on the privacy information in the user portrait stored locally; Receive personalized push notifications based on general and specific recommendation results.

Images