Subscription account content push method, device, equipment and storage medium

Abstract

The present application provides a method, device, equipment and storage medium for pushing content of a subscription account, which increases personalized push content when pushing messages to subscription account users. The method includes: obtaining the content of the operation record of the target user who follows the subscription account and multiple candidate contents matching the operation record, determining the content of the operation record and the feature vector of each candidate content, and determining the first push content with the greatest similarity to the content of the operation record from the multiple candidate contents through data processing of the feature vector, and sending the first push content to the terminal of the target user end. Since the push content of the subscription account is selected by the vector matching method, personalized push is achieved while also improving the diversity of the push content. By optimizing the push content, the user click rate can be improved.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to a method, device, equipment and storage medium for pushing content to a subscription account. Background Art

[0002] Subscription accounts provide a new way of disseminating information for media and individuals. Their main function is to convey information to users on terminal applications, such as pushing articles, activities, products and other content.

[0003] In the subscription account push scenario, taking the push of articles as an example, since the types and number of articles that can be pushed at a time are limited, most recommendation methods push the same article list to all users who follow the subscription account, and the selected articles are mainly recent popular articles. It can be seen that the current subscription account cannot achieve personalized recommendations. Summary of the invention

[0004] The embodiments of the present application provide a method, apparatus, device and storage medium for pushing content of a subscription account, so as to achieve personalized recommendation of subscription accounts and improve user experience.

[0005] A first aspect of an embodiment of the present application provides a method for pushing content of a subscription account, including:

[0006] Obtaining the content of the operation record of the target user who follows the subscription account, and multiple candidate contents matching the operation record;

[0007] Determine the content of the operation record and a feature vector of each candidate content;

[0008] Determine, according to the content of the operation record and the feature vector of each candidate content, a first pushed content from the plurality of candidate contents; the first pushed content is the candidate content having the greatest similarity to the content of the operation record;

[0009] The first pushed content is sent to a terminal of the target user.

[0010] In one embodiment of the present application, determining the content of the operation record and the feature vector of each candidate content includes:

[0011] The language model BERT is used to extract text features from the content of the operation record and each candidate content, so as to obtain a first feature vector of the content of the operation record and a second feature vector of each candidate content.

[0012] In one embodiment of the present application, the language model BERT is used to perform text feature extraction on the content of the operation record and each candidate content to obtain a first feature vector of the content of the operation record and a second feature vector of each candidate content, including:

[0013] Acquire first text information of the content of the operation record and second text information of each candidate content;

[0014] Using a preset dictionary of the language model BERT, convert the first text information into a first digital sequence, and convert the second text information of each candidate content into a second digital sequence;

[0015] The first digital sequence and the second digital sequence of each candidate content are respectively input into the language model BERT to obtain the first feature vector and the second feature vector of each candidate content.

[0016] In one embodiment of the present application, before converting the first text information into a first digital sequence using a preset dictionary of the language model BERT and converting the second text information of each candidate content into a second digital sequence, the method further includes:

[0017] The first text information and the second text information of each candidate content are respectively subjected to text preprocessing, wherein the text preprocessing at least includes word segmentation processing.

[0018] In one embodiment of the present application, determining the first pushed content from the plurality of candidate contents according to the content of the operation record and the feature vector of each candidate content includes:

[0019] Determine the cosine similarity between a first feature vector of the content of the operation record and a second feature vector of each candidate content;

[0020] The candidate content whose cosine similarity value is closest to 1 is used as the first pushed content.

[0021] In one embodiment of the present application, the first push content is located at the top of the push list of the push interface of the subscription account.

[0022] In one embodiment of the present application, the method further includes:

[0023] Sending second push content to the terminal of the target user, where the second push content is determined according to operation records of all users in the user group to which the target user belongs.

[0024] In one embodiment of the present application, the method further includes:

[0025] Obtaining the content of the operation record of each user in the same user group as the target user;

[0026] Determine a feature vector of the content of the operation record of each user in the user group;

[0027] Determine an average feature vector according to the feature vector of the content of the operation record of each user in the user group;

[0028] The second pushed content is determined from the multiple candidate contents of the user group according to the average feature vector and the feature vectors of the multiple candidate contents of the user group.

[0029] In one embodiment of the present application, obtaining the content of the operation record of each user in the same user group as the target user includes:

[0030] Acquire multiple other users who display the same first pushed content as the target user;

[0031] Classifying the target user and the multiple other users into the same user group;

[0032] The content of the operation record of each user in the same user group as the target user is obtained.

[0033] Optionally, the operation record includes at least one of a browsing record, an order record, and an additional purchase record.

[0034] A second aspect of an embodiment of the present application provides a content push device for a subscription account, including:

[0035] An acquisition module, used to acquire the content of the operation record of the target user who follows the subscription account, and multiple candidate contents matching the operation record;

[0036] A processing module, used to determine the content of the operation record and a feature vector of each candidate content;

[0037] Determine, according to the content of the operation record and the feature vector of each candidate content, a first pushed content from the plurality of candidate contents; the first pushed content is the candidate content having the greatest similarity to the content of the operation record;

[0038] A sending module is used to send the first pushed content to the terminal of the target user.

[0039] In one embodiment of the present application, the processing module is specifically used to:

[0040] The language model BERT is used to extract text features from the content of the operation record and each candidate content, so as to obtain a first feature vector of the content of the operation record and a second feature vector of each candidate content.

[0041] In one embodiment of the present application, the acquisition module is specifically used to:

[0042] Acquire first text information of the content of the operation record and second text information of each candidate content;

[0043] Processing module, specifically used for:

[0044] Using a preset dictionary of the language model BERT, convert the first text information into a first digital sequence, and convert the second text information of each candidate content into a second digital sequence;

[0045] The first digital sequence and the second digital sequence of each candidate content are respectively input into the language model BERT to obtain the first feature vector and the second feature vector of each candidate content.

[0046] In one embodiment of the present application, the processing module uses the preset dictionary of the language model BERT to convert the first text information into a first digital sequence, and before converting the second text information of each candidate content into a second digital sequence, is further used to:

[0047] The first text information and the second text information of each candidate content are respectively subjected to text preprocessing, wherein the text preprocessing at least includes word segmentation processing.

[0048] In one embodiment of the present application, the processing module is specifically used to:

[0049] Determine the cosine similarity between a first feature vector of the content of the operation record and a second feature vector of each candidate content;

[0050] The candidate content whose cosine similarity value is closest to 1 is used as the first pushed content.

[0051] In one embodiment of the present application, the first push content is located at the top of the push list of the push interface of the subscription account.

[0052] In one embodiment of the present application, the sending module is further used to: send second push content to the terminal of the target user, where the second push content is determined according to operation records of all users in the user group to which the target user belongs.

[0053] In one embodiment of the present application, the acquisition module is further used to:

[0054] Obtaining the content of the operation record of each user in the same user group as the target user;

[0055] The processing module is also used to:

[0056] Determine a feature vector of the content of the operation record of each user in the user group;

[0057] Determine an average feature vector according to the feature vector of the content of the operation record of each user in the user group;

[0058] The second pushed content is determined from the multiple candidate contents of the user group according to the average feature vector and the feature vectors of the multiple candidate contents of the user group.

[0059] In one embodiment of the present application, the acquisition module is specifically used to:

[0060] Acquire multiple other users who display the same first pushed content as the target user;

[0061] The processing module is further used to classify the target user and the multiple other users into the same user group;

[0062] The acquisition module is further used to acquire the content of the operation record of each user in the same user group as the target user.

[0063] Optionally, the operation record includes at least one of a browsing record, an order record, and an additional purchase record.

[0064] A third aspect of an embodiment of the present application provides an electronic device, including:

[0065] Memory;

[0066] Processor; and

[0067] Computer programs;

[0068] The computer program is stored in the memory and is configured to be executed by the processor to implement the method as described in any one of the first aspects.

[0069] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement a method as described in any one of the first aspects.

[0070] A fifth aspect of an embodiment of the present application provides a computer program product, including a computer program, which implements any method in the first aspect when executed by a processor.

[0071] The embodiments of the present application provide a method, device, equipment and storage medium for pushing content of a subscription account, which adds personalized push content when pushing messages to subscription account users. The method includes: obtaining the content of the operation record of the target user who follows the subscription account and multiple candidate contents matching the operation record, determining the feature vector of the content of the operation record and each candidate content, and determining the first push content with the greatest similarity to the content of the operation record from the multiple candidate contents through data processing of the feature vector, and sending the first push content to the terminal of the target user end. Since the push content of the subscription account is selected by the vector matching method, personalized push is achieved while also improving the diversification of the push content. By optimizing the push content, the user click rate can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0073] Figure 1 A schematic diagram of a scenario of a method for pushing content to a subscription account provided in an embodiment of the present application;

[0074] Figure 2 Schematic diagram of the process of the content push method of the subscription account provided in the embodiment of the present application Figure 1 ;

[0075] Figure 3 A flowchart of text feature extraction provided in an embodiment of the present application;

[0076] Figure 4 Schematic diagram of the process of the content push method of the subscription account provided in the embodiment of the present application Figure 2 ;

[0077] Figure 5 Schematic diagram of the process of the content push method of the subscription account provided in the embodiment of the present application Figure 3 ;

[0078] Figure 6 A schematic diagram of a push message of a push interface of a subscription account provided in an embodiment of the present application;

[0079] Figure 7 A schematic diagram of the structure of a content push device for a subscription account provided in an embodiment of the present application;

[0080] Figure 8 A hardware structure diagram of an electronic device provided in an embodiment of the present application.

[0081] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0082] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0083] It should be understood that the terms "including" and "having" and any variations thereof used herein are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0084] Subscription account push scenarios include new product launch push, popular product push, operational activity push, news push, special topic push, etc. Taking product push as an example, the most commonly used means of attracting users is to attract clicks by pushing titles containing keywords of user interest. However, due to restrictions on the style and type of content pushed each time and the frequency of pushes each month, subscription account pushes cannot be targeted and personalized for each subscription account fan, which is completely related to user interests.

[0085] In response to the above problems, an embodiment of the present application provides a method for pushing content to a subscription account, designs a push strategy that reflects personalization as much as possible, and maximizes the personalization of recommendations while satisfying various restrictions of the subscription account, thereby improving the click-through rate (or conversion rate) of users in the subscription account. The current subscription account can push a maximum of N links at a time, and N is usually less than 10, for example, 5. Taking into account the limit on the number of pushed links (or pushed articles / cards), the inventor designs the top link to correspond to the user's personalized content, and the remaining links to correspond to the mass content of the existing solution or the popular content in the user's group. In this way, targeted content recommendations for subscription accounts are achieved, thereby improving the user's click-through rate.

[0086] In addition, the above processing procedure is designed to determine the push content, and a text precise matching method can be used to determine the user's personalized content and the popular content in the user's group. However, the inventor also took into account the situation where Chinese characters have different expressions but the same semantics. In order to further improve the accuracy of the pushed content, a fuzzy matching method using the feature vector of the text is proposed to determine the above push content, thereby effectively solving the synonym matching problem and expanding the push content to more relevant content.

[0087] Before introducing the content push solution provided by this application, the application scenarios of the content push solution are briefly introduced.

[0088] For example, Figure 1 A scenario diagram of a method for pushing content to a subscription account provided in an embodiment of the present application, such as Figure 1 As shown, the scenario includes a terminal device 11, a first server 12, and at least one second server, for example Figure 1 The second servers 13 and 14 shown, the terminal device 11 is communicatively connected to the first server 12 , and the first server 12 is communicatively connected to the second servers 13 and 14 .

[0089] The terminal device 11 may be any electronic device with a display screen, including but not limited to a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.

[0090] The first server 12 may be a server that provides various network information, such as a server of an e-commerce platform, a server of a social platform, etc.

[0091] The second server 13 or 14 is a data processing server independent of the first server 12, and can be understood as a third-party server. For example, the second server 13 is a service platform for subscription number A, and the second server 14 is a service platform for subscription number B.

[0092] In one embodiment of the present application, the user accesses the first server 12 through the terminal device 11 to obtain various types of information. The first server 12 records the user's operation records. For example, the first server 12 is a server of a social platform, and the server records the user's browsing history; for another example, the first server 12 is a server of an e-commerce platform, and in addition to recording the user's browsing history, the server also records the user's order record, add-to-cart record, etc.

[0093] In one embodiment of the present application, the user subscribes to some subscription accounts on the first server 12 through the terminal device 11. The first server 12 regularly sends push content to the user through the subscription account. For example, the first server 12 is a server of a social platform, the user follows the public account of the social platform, and the server sends tweets of the public account to the user every day or every week; for another example, the first server 12 is a server of an e-commerce platform, the user follows the account of a store on the e-commerce platform, and the server sends new products of the store to the user every week or every month.

[0094] In one embodiment of the present application, the second server 13 or 14 sends the push content to the first server 12 by calling the interface provided by the first server 12, and then the first server 12 pushes the content to the user's terminal.

[0095] In the embodiment of the present application, the subscription account can be a public account, a service account, an enterprise account, etc., and the embodiment of the present application does not impose any restrictions on this. Among them, the subscription account is mainly used to convey information to users, and usually only one message can be sent per day. The service account is mainly for service interaction, such as banks, companies that provide service inquiries, number inquiries, etc., and usually several messages are sent every month. The enterprise account is mainly used for company content communication, and information can be pushed in the form of groups.

[0096] The technical solution of the present application is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0097] Figure 2 Schematic diagram of the process of the content push method of the subscription account provided in the embodiment of the present application Figure 1 The content push method provided in this embodiment can be applied to Figure 1 The second server 13 or 14 shown. Figure 2 As shown, the content push method specifically includes the following steps:

[0098] Step 101: Obtain the content of the operation record of the target user who follows the subscription account, and multiple candidate contents matching the operation record.

[0099] In this embodiment, the operation record of the target user includes at least one of a browsing record, an order record, and an additional purchase record of the target user in the subscription account that the target user follows or in the application of the subscription account.

[0100] Optionally, the operation record of the target user may refer to the operation record of the target user within a preset period of time before the current moment.

[0101] In one embodiment of the present application, the operation record of the target user who follows the subscription account can be obtained through the log file of the second server (the server of the subscription account) or the first server (the application server). The log file records the historical behavior of the user in the subscription account or application. Through the historical behavior of the user, the user's interest preferences can be understood, and the user's possible needs in the future can be predicted, such as the need to purchase products or services, and the need to read topic news.

[0102] The content of the above-mentioned operation record specifically refers to the product or information corresponding to the operation record. For example, if the operation record is a browsing record, the content of the browsing record includes the browsed products, articles, videos, etc.; for another example, if the operation record is an order record, the content of the order record includes the category of the ordered product, the quantity of the ordered product, the details of the ordered product, etc.; for another example, if the operation record is an add-to-cart record, the content of the add-to-cart record includes the category of the added-to-cart product, the quantity of the added-to-cart product, the details of the added-to-cart product, etc.

[0103] For example, taking an e-commerce subscription account as an example, user behaviors can be collected through the log files of the server of the e-commerce subscription account to obtain the content of the user's operation records. The collected user behaviors include: all product details in the user's shopping cart; the product collection of the same product category that the user has browsed more than three times in three days; the product collection of the same product category that the user has browsed three days in the past five days. Through the collected user behaviors, the specific content of the operation records corresponding to the user behaviors can be read to learn the products that the user is interested in recently.

[0104] In this embodiment, the multiple candidate contents matching the operation record may refer to multiple candidate products or multiple candidate information (including articles, videos, audio, etc.) having high similarity with the contents of the operation record.

[0105] For example, taking the candidate content as a candidate product set, the target user's preference for purchasing products can be determined through the specific content of the target user's operation record, thereby preliminarily screening out a recommended candidate product set from the product set, and taking all products in the candidate product set as multiple candidate content that matches the target user's operation record.

[0106] Step 102: Determine the content of the operation record and the feature vector of each candidate content.

[0107] In one embodiment of the present application, the language model BERT may be used to extract text features from the content of the operation record and each candidate content to obtain a first feature vector of the content of the operation record and a second feature vector of each candidate content.

[0108] Among them, BERT is a language model in the field of natural language processing (NLP), which can be used to encode text content and obtain a numerical feature vector corresponding to the text content.

[0109] For example, Figure 3 A flowchart of text feature extraction provided in the embodiment of the present application is shown in FIG. Figure 3 As shown, step 102 specifically includes the following steps:

[0110] Step 1021: Obtain first text information of the content of the operation record and second text information of each candidate content.

[0111] As can be seen from the above, the content of the operation record includes the product or information corresponding to the operation record, and obtaining the first text information of the content of the operation record can be understood as obtaining the text information including the title and / or the body of the product or information corresponding to the operation record. Similarly, obtaining the second text information of each candidate content can be understood as obtaining the text information including the title and / or the body of each candidate product or information.

[0112] Step 1022: Use the preset dictionary of the language model BERT to convert the first text information into a first digital sequence, and convert the second text information of each candidate content into a second digital sequence.

[0113] The preset dictionary of the language model BERT is the token-id dictionary, where token is the smallest processing unit of text in NLP. The preset dictionary includes the correspondence between text and numbers.

[0114] For example, the text is "Today's weather is really nice", and each word in this text corresponds to the numbers 16, 24, 24, 86, 55, and 70. For another example, the text is the English word "unbeautiful", and the word is segmented into three parts "un / beauti / ful" according to the prefix and suffix of the word, and the segmented phrases correspond to the numbers 2, 3, and 4 respectively.

[0115] The Chinese and / or English words in the text information are converted into machine-recognizable natural language processing text, that is, a digital sequence, through the preset dictionary of the language model BERT. The digital sequence is then input into the language model BERT in sequence to obtain the feature vector corresponding to the digital sequence.

[0116] Optionally, in some embodiments, before converting the first text information into a first digital sequence using the preset dictionary of the BERT language model and converting the second text information of each candidate content into a second digital sequence, it further includes: performing text preprocessing on the first text information and the second text information of each candidate content respectively, and the text preprocessing at least includes word segmentation processing.

[0117] The above example has shown a way of word segmentation processing for English text, that is, splitting a single English word according to prefixes and suffixes. If a word that does not exist in the dictionary is encountered, the word can be directly split according to letters. For example, the word "unknown" can be split into "un / k / n / o / w / n".

[0118] The above example has shown a way of word segmentation processing for Chinese text, that is, splitting according to the form of each single character, which is actually character splitting.

[0119] Optionally, in some embodiments, in addition to word segmentation processing, the text preprocessing further includes at least one of removing stop words, punctuation marks, and case conversion.

[0120] Among them, stop words refer to words such as "de", "le", "shi" in the text that have little effect on the construction of text features, and they can be removed from the text information according to the preset stop word dictionary. Punctuation marks contribute little to semantic features and should also be removed. Case conversion mainly involves English text. For example, "5G" is converted to "5g", "APPLE" is converted to "apple", etc., which is convenient for unified feature expression processing.

[0121] Optionally, in some embodiments, the text preprocessing further includes: adding start and end flags. According to the requirements of the BERT model input format, the two tokens "[CLS]" and "[SEP]" should be added before and after the text respectively as the start and end identifiers of the text, and the two added tokens are regarded as the size of two "characters".

[0122] For example, "[CLS] It's a nice day today [SEP]", the length of this text is 8. It should be noted that the preset dictionary also includes the numbers corresponding to the two added tokens. For example, [CLS] corresponds to the number 0, [SEP] corresponds to the number 1, then the digital sequence corresponding to the text "[CLS] It's a nice day today [SEP]" is [0 16 24 24 86 55 70 1].

[0123] Step 1023: Input the first digital sequence into the BERT language model to obtain a first feature vector.

[0124] Step 1024: Input the second digital sequence of each candidate content into the language model BERT to obtain the second feature vector of each candidate content.

[0125] The first feature vector and the second feature vector are text feature vectors, which represent the semantic features of the text. The above two steps are used to extract feature vectors, and the digital sequences obtained in step 1022 are sequentially input into the language model BERT accelerated by the GPU (graphics processor, i.e., graphics card). The features output by the last layer of the model are the required text feature vectors.

[0126] Step 103: Determine a first pushed content from a plurality of candidate contents according to the content of the operation record and the feature vector of each candidate content.

[0127] Step 104: Send the first push content to the terminal of the target user.

[0128] The first pushed content is the candidate content that has the greatest similarity to the content of the operation record.

[0129] In one embodiment of the present application, the first pushed content may be determined from a plurality of candidate contents according to the similarity between the content of the operation record and the feature vector of each candidate content.

[0130] Optionally, the similarity between two feature vectors may be determined by calculating cosine similarity.

[0131] Optionally, the similarity between two feature vectors may also be determined by using calculation formulas such as Euclidean distance and Manhattan distance.

[0132] In one embodiment of the present application, by determining the cosine similarity between the first feature vector of the operation record content and the second feature vector of each candidate content, the candidate content with the cosine similarity value closest to 1 is used as the first pushed content. The first pushed content is the pushed content most relevant to the content of the operation record of the target user.

[0133] In one embodiment of the present application, the first pushed content is located at the top of the push list of the push interface of the subscription account.

[0134] For example, if the user's operation records include the records of the products browsed by the user in the last three days, the first push content is the product that is the same as or most similar to the product browsed by the user. The product can be used as the "big picture" article in the subscription account push articles (i.e. the top of the push list or card), thereby increasing the user's click-through rate and conversion rate. Among them, the conversion rate includes search conversion rate, UV (abbreviation for unique visitor, which refers to people who visit and browse store pages through the Internet) order conversion rate, etc.

[0135] The above steps can be used to provide personalized recommendations to each user of the subscription account, including personalized recommendations of products, articles, videos, audios, activities, etc.

[0136] In this embodiment, considering the precise matching of text (or keyword matching), only the title text of products or information that absolutely matches the historical behavior of the target user can be matched, but it is impossible to match the title text expressed in different words but with the same semantics, and it is even more impossible to expand to other products or information that are related to the interests of the target user but not exactly the same. The present application proposes to replace the above text matching with fuzzy matching of vectors, which can effectively solve the problem of synonym matching, and can also expand the matching results to other related products or information by adjusting the similarity threshold, thereby improving the diversity of recommendations.

[0137] This embodiment provides a method for pushing content of a subscription account, which adds personalized push content when pushing messages to subscription account users. The method includes: obtaining the content of the operation record of the target user who follows the subscription account and multiple candidate contents matching the operation record, determining the feature vector of the content of the operation record and each candidate content, and determining the first push content with the greatest similarity to the content of the operation record from the multiple candidate contents through data processing of the feature vector, and sending the first push content to the terminal of the target user end. Since the push content of the subscription account is selected by the vector matching method, while realizing personalized push, the diversification of the push content is also improved. By optimizing the push content, the user click rate can be improved.

[0138] Considering that the number of pushed articles of a subscription account is usually limited, that is, a maximum of N articles can be sent at a time, for example, N is 5, and the top articles among the N articles can be determined through the above embodiment, then how are the remaining N-1 articles determined? In the existing solution, the N articles are all recent popular articles, and no targeted push is performed for users.

[0139] In order to increase the click-through rate of the target users on the remaining N-1 articles, you can consider grouping the users of the subscription account. For example, you can divide them into multiple user groups based on their interest preferences. You can match appropriate articles based on the operation records of all users in the user group, instead of sending the same remaining N-1 articles to all users. This can further improve the click-through rate of the remaining N-1 articles for users in the user group.

[0140] Figure 4 Schematic diagram of the process of the content push method of the subscription account provided in the embodiment of the present application Figure 2 The content push method provided in this embodiment can also be applied to Figure 1 The second server 13 or 14 shown. Figure 4 As shown, the content push method specifically includes the following steps:

[0141] Step 201: Obtain the content of the operation record of each user in the same user group as the target user.

[0142] In one embodiment of the present application, the user group to which the target user belongs may be determined based on the first pushed content that was most recently pushed to the target user.

[0143] Specifically, multiple other users who have the same first pushed content displayed to the target user are obtained, the target user and the multiple other users are classified into the same user group, and the content of the operation record of each user in the same user group as the target user is obtained.

[0144] Step 202: Determine a feature vector of the content of the operation record of each user in the user group.

[0145] In this embodiment, the feature vector of the content of the operation record of each user in the user group can be determined by referring to the implementation method of determining the feature vector of the content of the operation record of the target user in step 102 of the above embodiment, which will not be repeated here.

[0146] Step 203: Determine an average feature vector based on the feature vector of the content of the operation record of each user in the user group.

[0147] After determining the feature vector of the content of the operation record of each user in the user group, the average feature value of each feature dimension of the feature vector of all users in the user group is calculated respectively, so as to determine the average feature vector corresponding to the operation records of all users in the user group.

[0148] For example, the user group includes user 1, user 2, and user 3. The feature vector corresponding to the operation record of user 1 can be expressed as The feature vector corresponding to the operation record of user 2 can be expressed as The feature vector corresponding to the operation record of user 3 can be expressed as The average feature vector of all users in this user group can be expressed as:

[0149]

[0150] Step 204: Determine second pushed content from the plurality of candidate contents of the user group according to the average feature vector and the feature vectors of the plurality of candidate contents of the user group.

[0151] Step 205: Send the second push content to the terminal of the target user.

[0152] The second push content is determined according to the operation records of all users in the user group to which the target user belongs.

[0153] The number of second pushed content is at least one, that is, at least one second pushed content can be determined from multiple candidate contents of the user group. If the subscription account limits the number of pushed content to N, then the number of second pushed content is N-1, where N is a positive integer greater than or equal to 2.

[0154] In one embodiment of the present application, the second pushed content may be determined from the plurality of candidate contents of the user group according to the average feature vector and the similarity of the feature vectors of the plurality of candidate contents of the user group.

[0155] As in the above embodiment, the similarity between two feature vectors can be determined by calculating the cosine similarity.

[0156] In one embodiment of the present application, by determining the cosine similarity between the average feature vector and the feature vector of each candidate content of the user group, the first N-1 candidate contents are sorted according to the size of the cosine similarity, and the first N-1 candidate contents are taken as the second push content. The second push content is the push content that is most relevant to the content of the operation records of all users in the user group. Exemplarily, taking product recommendation as an example, the second push content is a set of products that most users in the user group may be interested in, and the product set includes N-1 recommended products.

[0157] In one embodiment of the present application, the second push content is located below the first push content at the top of the push interface of the subscription account, and is displayed in two forms: expanded or collapsed. The second push content displayed below the first push content is the "small picture" article in the subscription account push article.

[0158] It should be noted that users in the same user group will receive the same second pushed content, but users in the same user group generally receive different first pushed content.

[0159] This embodiment provides a method for pushing content of a subscription account. When pushing messages to a user of a subscription account, the operation status of other users in the user group to which the user belongs is considered, and the content that most users in the user group are interested in is pushed. The method includes: obtaining the content of the operation records of each user in the same user group as the target user, determining the average feature vector of the content of the operation records of all users in the user group, and selecting a suitable second push content from the multiple candidate contents of the user group by calculating the similarity of the average feature vector and the feature vectors of multiple candidate contents of the user group, thereby pushing the second push content to all users in the user group. Since the push content of users in the user group is selected by the vector matching method, it is possible to push popular content of similar user groups to users, which can improve the user click rate.

[0160] Figure 5 Schematic diagram of the process of the content push method of the subscription account provided in the embodiment of the present application Figure 3The content push method provided in this embodiment can also be applied to Figure 1 The second server 13 or 14 shown. Figure 5 As shown, the content push method specifically includes the following steps:

[0161] Step 301: Obtain the content of the operation record of the target user who follows the subscription account, and multiple candidate contents matching the operation record.

[0162] Step 302: Determine the content of the operation record and the feature vector of each candidate content.

[0163] Step 303: Determine a first pushed content from a plurality of candidate contents according to the content of the operation record and the feature vector of each candidate content.

[0164] Step 304: Obtain the content of the operation record of each user in the same user group as the target user.

[0165] Step 305: Determine a feature vector of the content of the operation record of each user in the user group.

[0166] Step 306: Determine an average feature vector based on the feature vector of the content of the operation record of each user in the user group.

[0167] Step 307: Determine second pushed content from the multiple candidate contents of the user group according to the average feature vector and the feature vectors of the multiple candidate contents of the user group.

[0168] Step 308: Send the first push content and the second push content to the terminal of the target user.

[0169] The specific implementation principles and technical effects of the above steps can be found in the above embodiments and will not be repeated here.

[0170] The subscription account content push method provided in this embodiment not only pushes a first push content that the target user is most interested in, but also pushes a second push content that is of interest to most users in a similar user group to the target user. While meeting the limit on the number of pushed articles, the above scheme takes into account the user's own interest preferences and the interest preferences of all users in the user group to which the user belongs, thereby maximizing the degree to which each user obtains personalized content recommendations. In addition, the language model is used to construct features for the content of user operation records and candidate content, which overcomes the problem that keyword matching cannot solve synonym matching, and can extend content recommendations to other similar but different content, increasing the diversity of recommended content.

[0171] For example, Figure 6 A schematic diagram of a push message of a push interface of a subscription account provided in an embodiment of the present application, such as Figure 6As shown, the push message of the push interface 401 includes a first push content and three second push contents, wherein the first push content is content directly related to the interest preference of the target user, and the second push content is content related to the overall interest preference of all users in the user group to which the target user belongs.

[0172] In a possible implementation, the three second push contents may be displayed in a tiled form on the push interface, such as Figure 6 (a).

[0173] In a possible implementation, the push interface only displays the first push content, and the three second push contents are simultaneously retracted (ie, hidden), such as Figure 6 (b).

[0174] In a possible implementation, the push interface displays the first push content and one second push content, and the remaining second push content is hidden, such as Figure 6 (c).

[0175] The user can view the retracted second push content by clicking the arrow.

[0176] The embodiment of the present application can divide the functional modules of the content push device of the subscription account according to the above method embodiment. For example, each functional module can be divided corresponding to each function, or two or more functions can be integrated into one processing module. The above integrated modules can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiment of the present application is schematic and is only a logical functional division. There may be other division methods in actual implementation. The following is an example of dividing each functional module corresponding to each function.

[0177] Figure 7 A schematic diagram of the structure of the content push device for a subscription account provided in an embodiment of the present application. Figure 7 As shown, the content push device 500 provided in this embodiment includes: an acquisition module 501 , a processing module 502 and a sending module 503 .

[0178] The acquisition module 501 is used to acquire the content of the operation record of the target user who follows the subscription account, and multiple candidate contents matching the operation record;

[0179] Processing module 502, used to determine the content of the operation record and the feature vector of each candidate content;

[0180] Determine, according to the content of the operation record and the feature vector of each candidate content, a first pushed content from the plurality of candidate contents; the first pushed content is the candidate content having the greatest similarity to the content of the operation record;

[0181] The sending module 503 is configured to send the first pushed content to the terminal of the target user.

[0182] In one embodiment of the present application, the processing module 502 is specifically configured to:

[0183] The language model BERT is used to extract text features from the content of the operation record and each candidate content, so as to obtain a first feature vector of the content of the operation record and a second feature vector of each candidate content.

[0184] In one embodiment of the present application, the acquisition module 501 is specifically used to:

[0185] Acquire first text information of the content of the operation record and second text information of each candidate content;

[0186] The processing module 502 is specifically configured to:

[0187] Using a preset dictionary of the language model BERT, convert the first text information into a first digital sequence, and convert the second text information of each candidate content into a second digital sequence;

[0188] The first digital sequence and the second digital sequence of each candidate content are respectively input into the language model BERT to obtain the first feature vector and the second feature vector of each candidate content.

[0189] In one embodiment of the present application, the processing module 502 uses the preset dictionary of the language model BERT to convert the first text information into a first digital sequence, and before converting the second text information of each candidate content into a second digital sequence, is further used to:

[0190] The first text information and the second text information of each candidate content are respectively subjected to text preprocessing, wherein the text preprocessing at least includes word segmentation processing.

[0191] In one embodiment of the present application, the processing module 502 is specifically configured to:

[0192] Determine the cosine similarity between a first feature vector of the content of the operation record and a second feature vector of each candidate content;

[0193] The candidate content whose cosine similarity value is closest to 1 is used as the first pushed content.

[0194] In one embodiment of the present application, the first push content is located at the top of the push list of the push interface of the subscription account.

[0195] In one embodiment of the present application, the sending module 503 is further used to: send a second push content to the terminal of the target user, where the second push content is determined according to the operation records of all users in the user group to which the target user belongs.

[0196] In one embodiment of the present application, the acquisition module 501 is further used to:

[0197] Obtaining the content of the operation record of each user in the same user group as the target user;

[0198] The processing module 502 is further configured to:

[0199] Determine a feature vector of the content of the operation record of each user in the user group;

[0200] Determine an average feature vector according to the feature vector of the content of the operation record of each user in the user group;

[0201] The second pushed content is determined from the multiple candidate contents of the user group according to the average feature vector and the feature vectors of the multiple candidate contents of the user group.

[0202] In one embodiment of the present application, the acquisition module 501 is specifically used to:

[0203] Acquire multiple other users who display the same first pushed content as the target user;

[0204] The processing module 502 is further configured to classify the target user and the multiple other users into the same user group;

[0205] The acquisition module 501 is further used to acquire the content of the operation record of each user in the same user group as the target user.

[0206] Optionally, the operation record includes at least one of a browsing record, an order record, and an additional purchase record.

[0207] The content push device for a subscription account provided in this embodiment can execute the technical solution of any of the above-mentioned method embodiments, and its implementation principles and technical effects are similar, which will not be repeated here.

[0208] Figure 8 The hardware structure diagram of the electronic device provided in the embodiment of the present application is as follows: Figure 8 As shown, the electronic device 600 provided in this embodiment includes:

[0209] Memory 601;

[0210] Processor 602; and

[0211] Computer programs;

[0212] Among them, the computer program is stored in the memory 601 and is configured to be executed by the processor 602 to implement the technical solution of any of the above-mentioned method embodiments. Its implementation principle and technical effect are similar and will not be repeated here.

[0213] Optionally, the memory 601 may be independent or integrated with the processor 602. When the memory 601 is a device independent of the processor 602, the electronic device 600 further includes a bus 604 for connecting the memory 601 and the processor 602.

[0214] The embodiment of the present application further provides a computer-readable storage medium on which a computer program is stored. The computer program is executed by the processor 602 to implement the technical solution in any of the aforementioned method embodiments.

[0215] An embodiment of the present application provides a computer program product, including a computer program, which implements the technical solution in any of the aforementioned method embodiments when executed by a processor.

[0216] An embodiment of the present application also provides a chip, including: a processing module and a communication interface, wherein the processing module can execute the technical solution in any of the aforementioned method embodiments.

[0217] Furthermore, the chip also includes a storage module (such as a memory), the storage module is used to store instructions, the processing module is used to execute the instructions stored in the storage module, and the execution of the instructions stored in the storage module enables the processing module to execute the technical solution in any of the aforementioned method embodiments.

[0218] It should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the invention may be directly implemented by a hardware processor or implemented by a combination of hardware and software modules in the processor.

[0219] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.

[0220] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0221] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0222] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic device.

[0223] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for pushing content of a subscription account, characterized in that: include: Obtaining the content of the operation record of the target user who follows the subscription account, and multiple candidate contents matching the operation record; the operation record includes at least one of the browsing record, order record, and additional purchase record of the target user in the subscription account or in the application of the subscription account; the content of the operation record refers to the product or information corresponding to the operation record; the subscription account is any one of a public account, a service account, and an enterprise account; Determine the content of the operation record and a feature vector of each candidate content; Determine a first push content from the plurality of candidate contents according to the content of the operation record and the feature vector of each candidate content; the first push content is the candidate content with the greatest similarity to the content of the operation record; the number of the first push content is 1; Sending the first push content to the terminal of the target user; the first push content is located at the top of the push list of the push interface of the subscription account; Sending a second push content to the terminal of the target user, where the second push content is determined based on the operation records of all users in the user group to which the target user belongs; the number of the second push content is N-1, where N is the number of push content of the subscription account this time, N is a positive integer greater than or equal to 2, and N is less than or equal to a preset value; the second push content is located below the first push content; The user group is determined according to the first pushed content pushed to the target user most recently; The determining the content of the operation record and the feature vector of each candidate content includes: Acquire first text information of the content of the operation record and second text information of each candidate content; Using a preset dictionary of the language model BERT, convert the first text information into a first digital sequence, and convert the second text information of each candidate content into a second digital sequence; The first digital sequence and the second digital sequence of each candidate content are respectively input into the language model BERT to obtain a first feature vector and a second feature vector of each candidate content.

2. The method according to claim 1, characterized in that Before converting the first text information into a first digital sequence using a preset dictionary of the language model BERT and converting the second text information of each candidate content into a second digital sequence, the method further includes: The first text information and the second text information of each candidate content are respectively subjected to text preprocessing, wherein the text preprocessing at least includes word segmentation processing.

3. The method according to claim 1 or 2, characterized in that: The determining the first pushed content from the plurality of candidate contents according to the content of the operation record and the feature vector of each candidate content includes: Determine the cosine similarity between a first feature vector of the content of the operation record and a second feature vector of each candidate content; The candidate content whose cosine similarity value is closest to 1 is used as the first pushed content.

4. The method according to claim 1, characterized in that: The method further comprises: Obtaining the content of the operation record of each user in the same user group as the target user; Determine a feature vector of the content of the operation record of each user in the user group; Determine an average feature vector according to the feature vector of the content of the operation record of each user in the user group; The second pushed content is determined from the multiple candidate contents of the user group according to the average feature vector and the feature vectors of the multiple candidate contents of the user group.

5. The method according to claim 4, characterized in that The obtaining of the operation record of each user in the same user group as the target user includes: Acquire multiple other users who display the same first pushed content as the target user; Classifying the target user and the multiple other users into the same user group; The content of the operation record of each user in the same user group as the target user is obtained.

6. A content push device for a subscription account, characterized in that: include: an acquisition module, for acquiring the content of the operation record of the target user who follows the subscription account, and a plurality of candidate contents matching the operation record; the operation record includes at least one of the browsing record, order record, and additional purchase record of the target user in the subscription account or in the application of the subscription account; the content of the operation record refers to the product or information corresponding to the operation record; the subscription account is any one of a public account, a service account, and an enterprise account; A processing module, used to determine the content of the operation record and a feature vector of each candidate content; Determine a first push content from the plurality of candidate contents according to the content of the operation record and the feature vector of each candidate content; the first push content is the candidate content with the greatest similarity to the content of the operation record; the number of the first push content is 1; A sending module, configured to send the first push content to the terminal of the target user; the first push content is located at the top of the push list of the push interface of the subscription account; The sending module is further used to send a second push content to the terminal of the target user, where the second push content is determined according to the operation records of all users in the user group to which the target user belongs; the number of the second push content is N-1, where N is the number of push content of the subscription account this time, N is a positive integer greater than or equal to 2, and N is less than or equal to a preset value; the second push content is located below the first push content; the user group is determined according to the first push content that was most recently pushed to the target user; The processing module is specifically used to obtain the first text information of the content of the operation record and the second text information of each candidate content; using the preset dictionary of the language model BERT, the first text information is converted into a first digital sequence, and the second text information of each candidate content is converted into a second digital sequence; the first digital sequence and the second digital sequence of each candidate content are respectively input into the language model BERT to obtain a first feature vector and a second feature vector of each candidate content.

7. An electronic device, characterized in that: include: Memory; processor; as well as Computer programs; The computer program is stored in the memory and is configured to be executed by the processor to implement the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is executed by a processor to implement the method according to any one of claims 1 to 5.

9. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 5 when being executed by a processor.

Images