A cold start personalized recommendation method and device for breaking the information cocoon problem
By identifying control accounts with similar user behavior during the cold start phase, generating a target interest matrix using their interest matrix, and adjusting the recommendation index, the problem of homogeneous recommended content for new users is solved, and the diversity of personalized recommendations is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NAT COMP NETWORK & INFORMATION SECURITY MANAGEMENT CENT
- Filing Date
- 2022-11-03
- Publication Date
- 2026-05-29
AI Technical Summary
During the initial cold start phase of an internet platform, newly registered users often experience data scarcity, leading to homogenized recommended content and a tendency to form information cocoons.
By identifying control accounts whose user behavior is similar to that of the target account to a threshold, the interest matrix of the target account is generated using the interest matrix of the control accounts. Combined with a preset content pool and quantification factors, the recommendation index is adjusted, and candidate recommendation data that matches the user's interests is pushed.
It solves the problem of homogeneous recommendations during the cold start phase, enriches the diversity of recommended content for users, and avoids the formation of information cocoons.
Smart Images
Figure CN115718835B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and in particular to a cold-start personalized recommendation method and apparatus for breaking the information cocoon problem. Background Technology
[0002] In recent years, with the rapid development of information technology, artificial intelligence technology has gradually matured in internet applications. Personalized content recommendation technology is an intelligent technology that uses user behavior data (such as browsing history, shopping records, followed information, and personal interest tags) to build a personalized user content recommendation model and predict user preferences. However, newly registered users generate relatively little data in the initial stages of using the internet, resulting in system-recommended content concentrated in a small range. This leads to homogenized content recommendations for users in the cold start phase, easily creating an information cocoon phenomenon.
[0003] There is currently no effective solution to the above problems. Summary of the Invention
[0004] This application provides a cold-start personalized recommendation method and apparatus to break the information cocoon problem, in order to solve the aforementioned technical problem that "the content recommended to users in the cold start stage tends to be homogenized".
[0005] According to one aspect of the embodiments of this application, this application provides a cold-start personalized recommendation method to overcome the information cocoon problem, comprising: when the target account is in the cold-start stage of its usage, identifying a reference account whose user behavior similarity to the target account reaches a first threshold; determining the target interest matrix of the target account using the reference account's reference interest matrix; determining a candidate recommendation set in a preset content pool using the target interest matrix, wherein the candidate recommendation set includes candidate recommendation data and the feature relevance between the candidate recommendation data and the target interest matrix; adjusting the weights of the feature relevance in the candidate recommendation set using a quantification factor to obtain a recommendation index for each candidate recommendation data; determining the candidate recommendation data whose recommendation index reaches a second threshold as the target recommendation data, and pushing the target recommendation data to the target account.
[0006] Optionally, determining the control accounts whose user behavior similarity to the target account reaches a first threshold includes: obtaining the target account's first interest data, and obtaining multiple user accounts and their second interest data from a user database, wherein the user database stores the interest data of user accounts that are not currently in the cold start phase during the cold start phase; generating a first interest matrix using the first interest data, and generating a second interest matrix using the second interest data; determining the matrix similarity between the first interest matrix and each of the second interest matrices to obtain the user behavior similarity between the target account and each user account; and determining the user accounts whose user behavior similarity reaches the first threshold as control accounts.
[0007] Optionally, determining the target interest matrix of the target account using the reference interest matrix of the reference account includes: obtaining the current interest data of the reference account and generating a reference interest matrix using the current interest data; and expanding the reference interest matrix using the first interest matrix of the target account to generate the target interest matrix.
[0008] Optionally, determining the candidate recommendation set in the preset content pool using the target interest matrix includes: obtaining the preset content pool, wherein the preset content pool includes a system content pool and a positive energy content pool; extracting the system candidate set and the positive energy candidate set from the system content pool and the positive energy content pool respectively using the target interest matrix; and determining the system candidate set and the positive energy candidate set as the candidate recommendation set.
[0009] Optionally, extracting the system candidate set and the positive energy candidate set from the system content pool and the positive energy content pool respectively using the target interest matrix includes: recalling a first content pool from the system content pool using the target interest matrix, and recalling a second content pool from the positive energy content pool using the target interest matrix; determining the first relevance between the data in the first content pool and the target interest matrix, and determining the second relevance between the data in the second content pool and the target interest matrix; extracting data from the first content pool whose first relevance reaches a third threshold to obtain the system candidate set, and extracting data from the second content pool whose second relevance reaches a fourth threshold to obtain the positive energy candidate set.
[0010] Optionally, the recommendation index for each candidate recommendation data is obtained by adjusting the weights of the feature relevance in the candidate recommendation set using quantification factors, including: determining the account quantification factor using the target browsing data of the target account and the total browsing data of all accounts on the platform where the target account is located, wherein the account quantification factor is used to characterize the proportion of the target account's target browsing data in the total browsing data, the quantification factor includes the account quantification factor and the system quantification factor, and the sum of the account quantification factor and the system quantification factor is a fixed value; multiplying the account quantification factor by the first relevance of the system candidate set to obtain the first recommendation index, and multiplying the system quantification factor by the second relevance of the positive energy candidate set to obtain the second recommendation index, wherein the feature relevance includes the first relevance and the second relevance, and the recommendation index includes the first recommendation index and the second recommendation index.
[0011] Optionally, determining the account quantification factor using the target browsing data of the target account and the total browsing data of all accounts on the platform to which the target account resides includes: obtaining the target browsing data of the target account and using the target browsing data to determine a first browsing frequency and a first category quantity; obtaining the total browsing data of all accounts and using the total browsing data to determine a second browsing frequency and a second category quantity; determining a first ratio of the first browsing frequency to the second browsing frequency and a second ratio of the first category quantity to the second category quantity; multiplying the first ratio and the second ratio, and determining the product as the account quantification factor.
[0012] According to another aspect of the embodiments of this application, this application provides a cold-start personalized recommendation device to break the information cocoon problem, comprising: an account determination module, used to determine a reference account whose user behavior similarity to the target account reaches a first threshold when the target account is in the cold-start stage of its usage; a matrix determination module, used to determine the target interest matrix of the target account using the reference interest matrix of the reference account; a recommendation set determination module, used to determine a candidate recommendation set in a preset content pool using the target interest matrix, wherein the candidate recommendation set includes candidate recommendation data and the feature relevance between the candidate recommendation data and the target interest matrix; an adjustment module, used to adjust the weight of the feature relevance in the candidate recommendation set using a quantification factor to obtain a recommendation index for each candidate recommendation data; and a data determination module, used to determine the candidate recommendation data whose recommendation index reaches a second threshold as target recommendation data and push the target recommendation data to the target account.
[0013] According to another aspect of the embodiments of this application, this application provides an electronic device, including a memory, a processor, a communication interface and a communication bus. The memory stores a computer program that can run on the processor. The memory and the processor communicate with each other through the communication bus and the communication interface. When the processor executes the computer program, it implements the steps of the above method.
[0014] According to another aspect of the embodiments of this application, this application also provides a computer-readable medium having processor-executable non-volatile program code that causes the processor to perform the above-described method.
[0015] Compared with related technologies, the technical solutions provided in this application have the following advantages:
[0016] This application proposes a cold-start personalized recommendation method to overcome the information cocoon problem. The method includes: identifying control accounts whose user behavior similarity to the target account reaches a first threshold when the target account is in the cold-start phase; determining the target account's target interest matrix using the control account's interest matrix; determining a candidate recommendation set from a preset content pool using the target interest matrix, wherein the candidate recommendation set includes candidate recommendation data and the feature relevance between the candidate recommendation data and the target interest matrix; adjusting the weights of the feature relevance in the candidate recommendation set using a quantification factor to obtain a recommendation index for each candidate recommendation data; and identifying the candidate recommendation data whose recommendation index reaches a second threshold as the target recommendation data and pushing the target recommendation data to the target account. By predicting the user account's interest matrix using control accounts, recalling the candidate recommendation set using the interest matrix, and finally selecting recommendation data based on the target account's browsing history, this method solves the problem of homogenized content recommendations for users in the cold-start phase. Attached Figure Description
[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0018] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.
[0019] Figure 1 A schematic diagram of the hardware environment for an optional cold-start personalized recommendation method for breaking the information cocoon problem provided in an embodiment of this application;
[0020] Figure 2 A flowchart illustrating an optional cold-start personalized recommendation method for breaking the information cocoon problem according to an embodiment of this application;
[0021] Figure 3 This is a block diagram of an optional cold-start personalized recommendation device for breaking the information cocoon problem according to an embodiment of this application;
[0022] Figure 4This is a schematic diagram of an optional electronic device structure provided in an embodiment of this application. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0024] In the following description, the use of suffixes such as "module," "part," or "unit" to denote elements is solely for the purpose of illustration and has no specific meaning in itself. Therefore, "module" and "part" may be used interchangeably.
[0025] In recent years, with the booming development of information technology, artificial intelligence technology has gradually matured in internet applications. Personalized content recommendation technology, in particular, is an intelligent technology that uses data such as user browsing history, shopping history, followed information, and personal interest tags to build personalized user content recommendation models and predict user preferences. Algorithm recommendation technology has gained widespread attention because it can accurately, quickly, and comprehensively help internet companies find potential target user groups. However, while bringing convenience to users, algorithm recommendation has also created some problems. The information cocoon problem is a prominent issue criticized by the internet industry. Algorithm recommendations constantly suggest content that users are interested in, limiting users' information selection space to a narrow range. Users become addicted to interest-based recommendations, and their attention becomes increasingly narrow. When users register new accounts or reset their recommendation settings on internet platforms, the content recommendation system obtains extremely sparse user data. This leads to users inadvertently clicking on certain types of content, especially vulgar or pandering content, causing the content recommendation system to frequently and indiscriminately push similar content, leading some audiences to become addicted to low-level entertainment. As a result, the content recommended to users during the cold start phase tends to be homogenized, which can easily lead to information cocoon phenomenon or even more serious consequences.
[0026] To address the problems mentioned in the background art, according to one aspect of the embodiments of this application, an embodiment of a cold-start personalized recommendation method for breaking the information cocoon problem is provided.
[0027] Optionally, in the embodiments of this application, the above-described cold-start personalized recommendation method for breaking the information cocoon problem can be applied to, for example... Figure 1 The hardware environment shown consists of terminal 101 and server 103. Figure 1As shown, server 103 is connected to terminal 101 via a network and can be used to provide services (data comparison service, data retrieval service, and data push service, etc.) to the terminal or the client installed on the terminal. Database 105 can be set up on the server or independently of the server to provide data storage services for server 103. The aforementioned network includes, but is not limited to, wide area network, metropolitan area network, or local area network. Terminal 101 includes, but is not limited to, PC, mobile phone, tablet computer, etc.
[0028] The cold-start personalized recommendation method for breaking the information cocoon problem in this application embodiment can be executed by server 103, or it can be jointly executed by server 103 and terminal 101, such as... Figure 2 As shown, it includes:
[0029] Step 201: If the target account is in the cold start phase of its usage, identify the control account whose user behavior similarity to the target account reaches the first threshold.
[0030] Step 203: Determine the target interest matrix of the target account using the comparison interest matrix of the comparison account;
[0031] Step 205: Use the target interest matrix to determine the candidate recommendation set in the preset content pool, wherein the candidate recommendation set includes candidate recommendation data and the feature correlation between the candidate recommendation data and the target interest matrix;
[0032] Step 207: Use quantification factors to adjust the weights of the feature relevance in the candidate recommendation set to obtain the recommendation index for each candidate recommendation data.
[0033] Step 209: Identify the candidate recommendation data that has reached the second threshold as the target recommendation data, and push the target recommendation data to the target account.
[0034] In fact, users can also choose to turn off algorithm recommendations directly to avoid creating an information cocoon and causing value bias. However, if users turn off the algorithm recommendation option directly, they will not be able to obtain suitable content that they are interested in and will not be able to enjoy personalized push services.
[0035] Therefore, it is necessary to design a cold start personalized content recommendation technology for users in the cold start environment (i.e., newly registered users or users after resetting the algorithm recommendation settings) to overcome the drawbacks of the information cocoon problem. After the recommendation system collects enough user interest data and passes through the cold start stage, users can continue to enjoy personalized recommendation services.
[0036] When a user completes registration or resets user recommendation settings on an internet platform, the personalized recommendation system has almost no user behavior data. This situation, where there is little or no user behavior data, is known as the cold start problem of the recommendation system.
[0037] This application proposes an embodiment of a personalized content recommendation method for cold starts aimed at breaking the information cocoon problem. Specifically, it includes: using a quantitative formula to distinguish whether a user is in the cold start phase; using collaborative filtering technology to obtain the interest points of user groups with similar interests and behaviors to the user in the cold start phase to compensate for the sparse interest matrix of the cold start user structure; and the recommendation system proactively introducing a pool of positive and uplifting content to enrich the recommended content for the user.
[0038] On the one hand, this application quantifies the cold start phase of a user's target account in order to optimize personalized content push technology for that user group.
[0039] Specifically, the method for determining whether a target account is in the cold start phase includes: obtaining target browsing data for the target account and using the target browsing data to determine the quantity of the first category; obtaining total browsing data for all accounts and using the total browsing data to determine the quantity of the second category; determining a second ratio between the quantity of the first category and the quantity of the second category; comparing the second ratio with a target threshold, and determining that the target account is in the cold start phase if the second ratio reaches the target threshold (e.g., 20%). The user can leave the cold start phase only when the category of the content viewed by the user is sufficiently representative of the user's interests.
[0040] On the other hand, it is necessary to enrich the attributes of sparse user data, expand user interests as much as possible, and improve the diversity of personalized recommendation content during the cold start phase.
[0041] Specifically, this application uses the current interest matrix of similar control accounts to predict the target interest matrix of the target account, then uses the target interest matrix to recall a candidate recommendation set from a preset content pool, and finally selects the recommendation data that the user is most likely to be interested in based on the user's preference weight.
[0042] As an optional embodiment, determining the control account whose user behavior similarity to the target account reaches a first threshold includes: obtaining first interest data of the target account, and obtaining multiple user accounts and second interest data of the user accounts from a user database, wherein the user database is used to store the interest data of user accounts that are not currently in the cold start phase during the cold start phase; generating a first interest matrix using the first interest data, and generating a second interest matrix using the second interest data; determining the matrix similarity between the first interest matrix and each of the second interest matrices to obtain the user behavior similarity between the target account and each user account; and determining the user account whose user behavior similarity reaches the first threshold as the control account.
[0043] In order to explore all user groups with similar interests to the target user's target account during the cold start phase, this application needs to pre-store the interest matrix of all users in the cold start phase, which requires additional storage overhead to assist in the calculation of user similarity groups during the cold start phase.
[0044] Specifically, the similarity of user behavior between the target account and other user accounts is obtained by calculating the matrix similarity between the target account's first interest matrix and the second interest matrix of other user accounts.
[0045] Optionally, the Pearson coefficient can be used to measure the similarity between users to obtain similar groups, and user behavior similarity in the cold start phase can be constructed to predict user interest matrix.
[0046] For example, user u is a user in the cold start phase, user j is a user who has passed the cold start phase, and the interest matrix of u and j in the cold start phase is calculated by user similarity to obtain the set of user groups with similar interests to u in the cold start phase.
[0047] The formula for calculating similarity is:
[0048]
[0049] in, R represents the average rating of user u for all content of interest. u，label This represents user u's rating of a single piece of content of interest (specific label). R represents the average rating of user j for all content of interest (all tags). j，label This represents user j's rating of a single piece of content of interest (specific label), while all_labels represents the collection of all content of interest (all labels).
[0050] After obtaining the similarity of user behavior between the target account and each user account, the user behavior similarity is ranked. If n similar accounts need to be selected, the similarity of the nth account is determined as the first threshold, and the user accounts whose user behavior similarity reaches the first threshold are determined as control accounts.
[0051] As an optional embodiment, determining the target interest matrix of the target account using the reference interest matrix of the reference account includes: obtaining the current interest data of the reference account and generating a reference interest matrix using the current interest data; and expanding the reference interest matrix using the first interest matrix of the target account to generate the target interest matrix.
[0052] Since the group of users with similar interests is not very large, the current interest data of one of the multiple comparison accounts is selected from the above comparison accounts in a dynamic and random manner, and then a comparison interest matrix is generated. Then, a matrix summation operation is performed with the first interest matrix of the target account to expand and form the target interest matrix. The target interest matrix is temporarily set as the interest matrix of the target user in the cold start phase.
[0053] As an optional embodiment, determining the candidate recommendation set in a preset content pool using a target interest matrix includes: obtaining a preset content pool, wherein the preset content pool includes a system content pool and a positive energy content pool; extracting a system candidate set and a positive energy candidate set from the system content pool and the positive energy content pool respectively using the target interest matrix; and determining the system candidate set and the positive energy candidate set as the candidate recommendation set.
[0054] The system content pool includes data from the recommendation system, while the positive energy content pool includes positive and uplifting data.
[0055] The target interest matrix is used to extract system candidate sets and positive energy candidate sets from the system content pool and positive energy content pool, respectively. In fact, this means recalling the datasets related to the target interest matrix from the system content pool and positive energy content pool to obtain the corresponding sets. The following will explain this step in more detail.
[0056] As an optional embodiment, extracting system candidate sets and positive energy candidate sets from the system content pool and positive energy content pool respectively using the target interest matrix includes: recalling a first content pool from the system content pool using the target interest matrix, and recalling a second content pool from the positive energy content pool using the target interest matrix; determining a first correlation between the data in the first content pool and the target interest matrix, and determining a second correlation between the data in the second content pool and the target interest matrix; extracting data from the first content pool whose first correlation reaches a third threshold to obtain the system candidate set, and extracting data from the second content pool whose second correlation reaches a fourth threshold to obtain the positive energy candidate set.
[0057] Optionally, if the target interest matrix includes interest tags, then the tags can be used to recall data that matches the interest tags from the system content pool and the positive energy content pool; if the target interest matrix does not include interest tags, then the data that appears most frequently in the target interest matrix can be used as the basis for recall.
[0058] In addition, the recalled data is sorted and refined to obtain new interest data with higher relevance. Then, the interest data is sorted according to the relevance between the interest data and the target interest matrix (including the first relevance and the second relevance). The data with the first relevance reaching the third threshold is determined as the system candidate set, and the data with the second relevance reaching the fourth threshold is determined as the positive energy candidate set. The third threshold and the fourth threshold can be set according to the situation, and this application does not make specific limitations.
[0059] Using a target interest matrix as the matching basis, highly relevant data is selected from a pre-set content pool, ensuring that the selected data maintains a certain correlation with the target account's interests. The candidate recommendation set includes both the system candidate set and the positive energy candidate set, enriching the categories of recommended data.
[0060] As an optional embodiment, the recommendation index for each candidate recommendation data is obtained by adjusting the weights of the feature relevance in the candidate recommendation set using a quantification factor. This includes: determining an account quantification factor using the target browsing data of the target account and the total browsing data of all accounts on the platform to which the target account belongs. The account quantification factor is used to characterize the proportion of the target account's target browsing data in the total browsing data. The quantification factor includes an account quantification factor and a system quantification factor, and the sum of the account quantification factor and the system quantification factor is a fixed value. The account quantification factor is multiplied by the first relevance of the system candidate set to obtain a first recommendation index, and the system quantification factor is multiplied by the second relevance of the positive energy candidate set to obtain a second recommendation index. The feature relevance includes the first relevance and the second relevance, and the recommendation index includes the first recommendation index and the second recommendation index.
[0061] Optionally, if factor(u) is defined as the account quantification factor, then the system quantification factor is (1-factor(u)). The account quantification factor can be simply understood as the weight of the data generated by the subjective click behavior of the target user in the total data of the system or the sparsity of the interest matrix, i.e., subjective weight. Then the system quantification factor can be understood as objective weight.
[0062] The first recommendation index is obtained by multiplying the account quantification factor by the first relevance of the system candidate set. The first recommendation index is the user preference of the data in the system candidate set after weight adjustment. The second recommendation index is obtained by multiplying the system quantification factor by the second relevance of the positive energy candidate set. The second recommendation index is the fit between the data in the positive energy candidate set and the user after weight adjustment.
[0063] Preferably, after obtaining the recommendation index of each candidate recommendation data, the data is recommended to the target account in order of the recommendation index. The first and second recommendation indices can be sorted separately, resulting in separate system-recommended content and positive content being pushed out, ensuring a diverse range of recommended content for the user. Alternatively, the first and second recommendation indices can be sorted together, and then content is recommended to the target user in descending order of recommendation index, ensuring that the recommended content received by the user is most suitable for their preferences.
[0064] As an optional embodiment, determining the account quantification factor using the target browsing data of the target account and the total browsing data of all accounts on the platform to which the target account belongs includes: obtaining the target browsing data of the target account and using the target browsing data to determine a first browsing frequency and a first category quantity; obtaining the total browsing data of all accounts and using the total browsing data to determine a second browsing frequency and a second category quantity; determining a first ratio of the first browsing frequency to the second browsing frequency and determining a second ratio of the first category quantity to the second category quantity; multiplying the first ratio and the second ratio, and determining the product as the account quantification factor.
[0065] For example, the formula for calculating the account quantification factor factor(u) is:
[0066]
[0067] It should be noted that the second ratio needs to be calculated at the beginning of the embodiments of this application, because the second ratio is also used to determine whether the usage phase of the target account is the cold start phase.
[0068] This application also has the following beneficial effects: ① Application of collaborative filtering technology: By adopting user-based collaborative filtering technology, another group of users with the same interests and hobbies can be explored to solve the problem of sparse user interest information during cold starts and to construct a more compact candidate matrix of user interests during cold starts; ② Application of natural language processing technology: Word2vec, text edit distance, NLPIR and Viterbi algorithm, etc., by calculating the text edit distance between the user's interest set text and the positive energy content pool, positive energy content can be recommended to users in a personalized way, improving the diversity of content displayed to users, breaking the information cocoon, and preventing users from falling into vulgar interests and hobbies; ③ The algorithm has many adjustable parameters that can be set according to needs, low internal coupling, and good portability.
[0069] This application proposes a cold-start personalized recommendation method to overcome the information cocoon problem. The method includes: identifying control accounts whose user behavior similarity to the target account reaches a first threshold when the target account is in the cold-start phase; determining the target account's target interest matrix using the control account's interest matrix; determining a candidate recommendation set from a preset content pool using the target interest matrix, wherein the candidate recommendation set includes candidate recommendation data and the feature relevance between the candidate recommendation data and the target interest matrix; adjusting the weights of the feature relevance in the candidate recommendation set using a quantification factor to obtain a recommendation index for each candidate recommendation data; and identifying the candidate recommendation data whose recommendation index reaches a second threshold as the target recommendation data and pushing the target recommendation data to the target account. By predicting the user account's interest matrix using control accounts, recalling the candidate recommendation set using the interest matrix, and finally selecting recommendation data based on the target account's browsing history, this method solves the problem of homogenized content recommendations for users in the cold-start phase.
[0070] According to another aspect of the embodiments of this application, this application provides a cold-start personalized recommendation device to break the information cocoon problem, such as... Figure 3 As shown, it includes:
[0071] The account determination module 302 is used to determine the reference account whose user behavior similarity to the target account reaches a first threshold when the target account is in the cold start phase of its usage.
[0072] The matrix determination module 304 is used to determine the target interest matrix of the target account using the comparison interest matrix of the comparison account;
[0073] The recommendation set determination module 306 is used to determine a candidate recommendation set in a preset content pool using the target interest matrix. The candidate recommendation set includes candidate recommendation data and the feature correlation between the candidate recommendation data and the target interest matrix.
[0074] The adjustment module 308 is used to adjust the weight of the feature relevance in the candidate recommendation set using a quantification factor to obtain the recommendation index of each candidate recommendation data.
[0075] The data determination module 310 is used to determine the candidate recommendation data that has reached the second threshold as the target recommendation data and push the target recommendation data to the target account.
[0076] It should be noted that the account determination module 302 in this embodiment can be used to execute step 201 in this application embodiment, the matrix determination module 304 in this embodiment can be used to execute step 203 in this application embodiment, the recommendation set determination module 306 in this embodiment can be used to execute step 205 in this application embodiment, the adjustment module 308 in this embodiment can be used to execute step 207 in this application embodiment, and the data determination module 310 in this embodiment can be used to execute step 209 in this application embodiment.
[0077] Optionally, the account determination module 302 is further configured to obtain first interest data of the target account, and obtain multiple user accounts and second interest data of the user accounts from the user database, wherein the user database is used to store the interest data of user accounts that are not currently in the cold start phase during the cold start phase; generate a first interest matrix using the first interest data, and generate a second interest matrix using the second interest data; determine the matrix similarity between the first interest matrix and each second interest matrix to obtain the user behavior similarity between the target account and each user account; and determine the user accounts whose user behavior similarity reaches a first threshold as control accounts.
[0078] Optionally, the matrix determination module 304 is further configured to obtain the current interest data of the reference account and generate a reference interest matrix using the current interest data; and to expand the reference interest matrix using the first interest matrix of the target account to generate the target interest matrix.
[0079] Optionally, the recommendation set determination module 306 includes:
[0080] The acquisition unit is used to acquire a preset content pool, which includes a system content pool and a positive energy content pool.
[0081] The extraction unit is used to extract system candidate sets and positive energy candidate sets from the system content pool and positive energy content pool respectively using the target interest matrix; and to determine the system candidate sets and positive energy candidate sets as candidate recommendation sets.
[0082] Optionally, the extraction unit is further configured to recall a first content pool from the system content pool using the target interest matrix, and to recall a second content pool from the positive energy content pool using the target interest matrix; determine a first correlation between the data in the first content pool and the target interest matrix, and determine a second correlation between the data in the second content pool and the target interest matrix; extract data from the first content pool whose first correlation reaches a third threshold to obtain a system candidate set, and extract data from the second content pool whose second correlation reaches a fourth threshold to obtain a positive energy candidate set.
[0083] Optionally, the adjustment module 308 includes:
[0084] The determination unit is used to determine the account quantification factor by using the target browsing data of the target account and the total browsing data of all accounts on the platform where the target account is located. The account quantification factor is used to characterize the proportion of the target account's target browsing data in the total browsing data. The quantification factor includes the account quantification factor and the system quantification factor, and the sum of the account quantification factor and the system quantification factor is a fixed value.
[0085] The multiplication unit is used to multiply the account quantification factor with the first relevance of the system candidate set to obtain the first recommendation index, and to multiply the system quantification factor with the second relevance of the positive energy candidate set to obtain the second recommendation index. The feature relevance includes the first relevance and the second relevance, and the recommendation index includes the first recommendation index and the second recommendation index.
[0086] Optionally, the determining unit is further configured to acquire target browsing data of the target account, and use the target browsing data to determine a first browsing frequency and a first category quantity; acquire total browsing data of all accounts, and use the total browsing data to determine a second browsing frequency and a second category quantity; determine a first ratio of the first browsing frequency to the second browsing frequency, and determine a second ratio of the first category quantity to the second category quantity; multiply the first ratio and the second ratio, and determine the resulting product as an account quantification factor.
[0087] It should be noted that the examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should also be noted that the above modules, as part of a device, can operate in environments such as... Figure 1 The hardware environment shown.
[0088] According to another aspect of the embodiments of this application, this application provides an electronic device, such as... Figure 4 As shown, the system includes a memory 401, a processor 403, a communication interface 405, and a communication bus 407. The memory 401 stores a computer program that can run on the processor 403. The memory 401 and the processor 403 communicate through the communication interface 405 and the communication bus 407. When the processor 403 executes the computer program, it implements the steps of the above method.
[0089] The memory and processor in the aforementioned electronic devices communicate with each other via a communication bus and a communication interface. The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc.
[0090] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.
[0091] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0092] According to another aspect of the embodiments of this application, a computer-readable medium having processor-executable non-volatile program code is also provided.
[0093] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.
[0094] In specific implementation, the embodiments of this application can be referred to the above embodiments and have corresponding technical effects.
[0095] It is understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described herein, or combinations thereof.
[0096] For software implementation, the techniques described herein can be implemented by units that perform the functions described herein. The software code can be stored in memory and executed by a processor. The memory can be implemented in the processor or external to the processor.
[0097] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0098] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0099] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.
[0100] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0101] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.
[0102] If the aforementioned function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks. It should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0103] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.
Claims
1. A cold-start personalized recommendation method to break the information cocoon problem, characterized in that, include: When the target account is in the cold start phase of its usage, identify a reference account whose user behavior similarity to the target account reaches a first threshold. The target interest matrix of the target account is determined using the comparison interest matrix of the comparison account; The target interest matrix is used to determine a candidate recommendation set in a preset content pool. The candidate recommendation set includes candidate recommendation data and the feature correlation between the candidate recommendation data and the target interest matrix. The candidate recommendation data includes data from the system content pool and data from the positive energy pool that are related to the target interest matrix. The step of determining the candidate recommendation set in the preset content pool using the target interest matrix includes: obtaining the preset content pool, wherein the preset content pool includes a system content pool and a positive energy content pool; extracting a system candidate set and a positive energy candidate set from the system content pool and the positive energy content pool respectively using the target interest matrix; and determining the system candidate set and the positive energy candidate set as the candidate recommendation set. The step of extracting system candidate sets and positive energy candidate sets from the system content pool and the positive energy content pool respectively using the target interest matrix includes: recalling a first content pool from the system content pool using the target interest matrix, and recalling a second content pool from the positive energy content pool using the target interest matrix; determining a first relevance between the data in the first content pool and the target interest matrix, and determining a second relevance between the data in the second content pool and the target interest matrix; extracting data from the first content pool whose first relevance reaches a third threshold to obtain the system candidate set, and extracting data from the second content pool whose second relevance reaches a fourth threshold to obtain the positive energy candidate set; The recommendation index for each candidate recommendation data is obtained by adjusting the weights of the feature relevance in the candidate recommendation set using a quantification factor, including: determining an account quantification factor using the target browsing data of the target account and the total browsing data of all accounts on the platform where the target account is located, wherein the account quantification factor is used to characterize the proportion of the target browsing data of the target account in the total browsing data, the quantification factor includes the account quantification factor and the system quantification factor, and the sum of the account quantification factor and the system quantification factor is a fixed value; multiplying the account quantification factor by the first relevance of the system candidate set to obtain a first recommendation index, and multiplying the system quantification factor by the second relevance of the positive energy candidate set to obtain a second recommendation index, wherein the feature relevance includes the first relevance and the second relevance, and the recommendation index includes the first recommendation index and the second recommendation index; The candidate recommendation data that reaches the second threshold is identified as the target recommendation data, and the target recommendation data is pushed to the target account.
2. The method according to claim 1, characterized in that, The reference accounts whose user behavior similarity to the target account reaches a first threshold include: The system obtains the first interest data of the target account and retrieves multiple user accounts and the second interest data of the user accounts from the user database. The user database is used to store the interest data of the user accounts that are not currently in the cold start phase during the cold start phase. A first interest matrix is generated using the first interest data, and a second interest matrix is generated using the second interest data; Determine the matrix similarity between the first interest matrix and each of the second interest matrices to obtain the user behavior similarity between the target account and each of the user accounts; The user accounts whose user behavior similarity reaches the first threshold are identified as the reference accounts.
3. The method according to claim 2, characterized in that, The step of determining the target interest matrix of the target account using the comparison interest matrix of the comparison account includes: Obtain the current interest data of the comparison account, and use the current interest data to generate the comparison interest matrix; The target interest matrix is generated by expanding the comparison interest matrix using the first interest matrix of the target account.
4. The method according to claim 1, characterized in that, The process of determining the account quantification factor using the target browsing data of the target account and the total browsing data of all accounts on the platform to which the target account belongs includes: Obtain the target browsing data of the target account, and use the target browsing data to determine a first browsing frequency and a first category quantity; and Obtain the total browsing data of all accounts, and use the total browsing data to determine the second browsing frequency and the number of the second category; Determine a first ratio of the first browsing frequency to the second browsing frequency, and determine a second ratio of the number of the first category to the number of the second category; Multiply the first ratio by the second ratio, and determine the resulting product as the account quantification factor.
5. A cold-start personalized recommendation device for breaking the information cocoon problem, characterized in that, include: The account determination module is used to determine a reference account whose user behavior similarity to the target account reaches a first threshold when the target account is in the cold start phase of its usage. A matrix determination module is used to determine the target interest matrix of the target account using the reference interest matrix of the reference account; The recommendation set determination module is used to determine a candidate recommendation set in a preset content pool using the target interest matrix. The candidate recommendation set includes candidate recommendation data and the feature correlation between the candidate recommendation data and the target interest matrix. The candidate recommendation data includes data from the system content pool and data from the positive energy pool that are related to the target interest matrix. An adjustment module is used to adjust the weights of the feature relevance in the candidate recommendation set using a quantification factor to obtain a recommendation index for each candidate recommendation data. This includes: determining an account quantification factor using the target browsing data of the target account and the total browsing data of all accounts on the platform where the target account resides; wherein the account quantification factor characterizes the proportion of the target browsing data of the target account in the total browsing data; the quantification factor includes the account quantification factor and a system quantification factor, the sum of which is a fixed value; multiplying the account quantification factor by a first relevance of the system candidate set to obtain a first recommendation index; and multiplying the system quantification factor by a second relevance of the positive energy candidate set to obtain a second recommendation index; wherein the feature relevance includes the first relevance and the second relevance, and the recommendation index includes the first recommendation index and the second recommendation index. The data determination module is used to determine the candidate recommendation data whose recommendation index reaches the second threshold as the target recommendation data, and push the target recommendation data to the target account; The recommendation set determination module is further configured to obtain the preset content pool, wherein the preset content pool includes a system content pool and a positive energy content pool; extract system candidate sets and positive energy candidate sets from the system content pool and the positive energy content pool respectively using the target interest matrix; and determine the system candidate sets and positive energy candidate sets as the candidate recommendation sets; The recommendation set determination module is further configured to recall a first content pool from the system content pool using the target interest matrix, and recall a second content pool from the positive energy content pool using the target interest matrix; determine a first relevance between the data in the first content pool and the target interest matrix, and determine a second relevance between the data in the second content pool and the target interest matrix; extract data from the first content pool whose first relevance reaches a third threshold to obtain the system candidate set, and extract data from the second content pool whose second relevance reaches a fourth threshold to obtain the positive energy candidate set.
6. An electronic device comprising a memory, a processor, a communication interface, and a communication bus, wherein the memory stores a computer program executable on the processor, and the memory and the processor communicate via the communication bus and the communication interface, characterized in that... When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.
7. A computer-readable medium having processor-executable non-volatile program code, characterized in that, The program code causes the processor to execute the method of any one of claims 1 to 4.