Index recommendation method and device, electronic equipment and storage medium

Abstract

The invention discloses an index recommendation method and device, electronic equipment and a storage medium, and the method comprises the steps: determining user identification information and a current target operation region of a user, calling a target index recommendation model corresponding to the target operation region, and generating a corresponding recommendation index for the user identification information based on the target index recommendation model, different operation areas generate corresponding recommendation indexes through different index recommendation models, and contents corresponding to the recommendation indexes are output and displayed. In combination with multi-dimensional user data (user data corresponding to different operation areas) and business scene information, more accurate index recommendation is realized.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, and in particular to a method and device for index recommendation, an electronic device, and a storage medium. Background Art

[0002] Existing technologies mainly select the most likely indicator to recommend to users based on Pearson similarity or Yuxuan similarity. Results can only be mined based on a single factor (indicator or user) and recommended to users. This indicator recommendation method may cause a lag between the recommendation results and user preferences, making the recommendation results inaccurate. Summary of the invention

[0003] The present disclosure provides a method, device, electronic device and storage medium for index recommendation, the main purpose of which is to solve the problem of achieving accurate index recommendation.

[0004] According to a first aspect of the present disclosure, a method for index recommendation is provided, comprising:

[0005] Determine user identification information and the user's current target operation area;

[0006] Calling a target indicator recommendation model corresponding to the target operation area, and generating a corresponding recommendation indicator for the user identification information based on the target indicator recommendation model, wherein different operation areas generate corresponding recommendation indicators through different indicator recommendation models;

[0007] The content corresponding to the recommendation indicator is output and displayed.

[0008] In some embodiments of the present application, before calling the target indicator recommendation model corresponding to the target operation area and generating the corresponding recommendation indicator for the user identification information based on the target indicator recommendation model, the method further includes:

[0009] Analyze the user operation history data to obtain the operation history log and the associated user basic information and indicator basic information;

[0010] Constructing a dynamic factor matrix according to the operation history log, the user basic information and the indicator basic information;

[0011] Performing dimensionality reduction processing on the dynamic factor matrix to obtain a local factor matrix;

[0012] Constructing different indicator recommendation models according to the local factor matrix, wherein the indicator recommendation model includes a first indicator recommendation model, a second indicator recommendation model, a third indicator recommendation model and a fourth indicator recommendation model;

[0013] Different indicator recommendation models are trained respectively to obtain trained indicator recommendation models.

[0014] In some embodiments of the present application, constructing a dynamic factor matrix according to the operation history log, the user basic information and the indicator basic information includes:

[0015] Calculate the attention level of each user to the indicator according to the operation history log, the user basic information and the indicator basic information;

[0016] Constructing the dynamic factor matrix according to each user's attention to the indicator;

[0017] The dynamic factor matrix is ​​decomposed into the product of the user and hidden trait matrix, and the hidden trait and indicator matrix.

[0018] In some embodiments of the present application, the calculating, according to the operation history log, the user basic information and the indicator basic information, the degree of attention of each user to the indicator includes:

[0019] Calculate the user's score for a single indicator and the average of the user's scores for all indicators based on the user basic information and the indicator basic information;

[0020] The cosine similarity calculation is performed based on the user's score for a single indicator and the average value of the user's score for all indicators to obtain the degree of attention of each user to the indicator.

[0021] In some embodiments of the present application, performing dimensionality reduction processing on the dynamic factor matrix to obtain a local factor matrix includes:

[0022] Performing feature fusion processing on the user and hidden trait matrix, and the hidden trait and indicator matrix to obtain a fused matrix;

[0023] The FunkSVD algorithm is called to optimize the fused matrix to obtain a local factor matrix.

[0024] In some embodiments of the present application, constructing different indicator recommendation models according to the local factor matrix includes:

[0025] Based on the similarities between users in the operation history log, construct the first indicator recommendation model;

[0026] Building the second indicator recommendation model based on the operation log data of the operation type of concern in the operation history log;

[0027] Based on the operation log data in the operation history log whose operation type is search, construct the third indicator recommendation model;

[0028] Based on the indicator information of the operation type of the current operation in the operation history log, the fourth indicator recommendation model is constructed according to the indicator change trend, correlation analysis, and influencing factor identification.

[0029] In some embodiments of the present application, respectively training different indicator recommendation models to obtain trained indicator recommendation models includes:

[0030] The local factor matrix is ​​used as the input of the indicator recommendation model, and the SVD++ algorithm is used for training to obtain a trained indicator recommendation model.

[0031] According to a second aspect of the present disclosure, there is provided a device for index recommendation, comprising:

[0032] A determination unit, used to determine user identification information and the user's current target operation area;

[0033] A first processing unit is used to call a target indicator recommendation model corresponding to the target operation area, and generate a corresponding recommendation indicator for the user identification information based on the target indicator recommendation model, wherein different operation areas generate corresponding recommendation indicators through different indicator recommendation models;

[0034] The display unit is used to output and display the content corresponding to the recommendation indicator.

[0035] In some embodiments of the present application, the device further comprises:

[0036] A parsing unit, configured to parse the user operation history data to obtain the operation history log and the respectively associated user basic information and indicator basic information before calling the target indicator recommendation model corresponding to the target operation area and generating the corresponding recommendation indicator for the user identification information based on the target indicator recommendation model;

[0037] A first construction unit is used to construct a dynamic factor matrix according to the operation history log, the user basic information and the indicator basic information;

[0038] A second processing unit is used to perform dimensionality reduction processing on the dynamic factor matrix to obtain a local factor matrix;

[0039] A second construction unit is used to construct different indicator recommendation models according to the local factor matrix, wherein the indicator recommendation model includes a first indicator recommendation model, a second indicator recommendation model, a third indicator recommendation model and a fourth indicator recommendation model;

[0040] The training unit is used to train different indicator recommendation models respectively to obtain trained indicator recommendation models.

[0041] In some embodiments of the present application, the first building block includes:

[0042] A calculation module, used to calculate the degree of attention of each user to the indicator according to the operation history log, the user basic information and the indicator basic information;

[0043] A construction module, used for constructing the dynamic factor matrix according to the degree of attention paid by each user to the indicator;

[0044] The decomposition module is used to decompose the dynamic factor matrix into the product of the user and hidden trait matrix and the hidden trait and indicator matrix.

[0045] In some embodiments of the present application, the calculation module is further used for:

[0046] Calculate the user's score for a single indicator and the average of the user's scores for all indicators based on the user's basic information and the indicator basic information;

[0047] The cosine similarity calculation is performed based on the user's score for a single indicator and the average value of the user's score for all indicators to obtain the degree of attention of each user to the indicator.

[0048] In some embodiments of the present application, the building block is further used for:

[0049] Perform feature fusion processing on the user and hidden trait matrix, and the hidden trait and indicator matrix to obtain a fused matrix;

[0050] The FunkSVD algorithm is called to optimize the fused matrix to obtain a local factor matrix.

[0051] In some embodiments of the present application, the second building unit is further used for:

[0052] Based on the similarities between users in the operation history log, construct the first indicator recommendation model;

[0053] Building the second indicator recommendation model based on the operation log data of the operation type of concern in the operation history log;

[0054] Based on the operation log data in the operation history log whose operation type is search, construct the third indicator recommendation model;

[0055] Based on the indicator information of the operation type of the current operation in the operation history log, the fourth indicator recommendation model is constructed according to the indicator change trend, correlation analysis, and influencing factor identification.

[0056] In some embodiments of the present application, the training unit is further used to:

[0057] The local factor matrix is ​​used as the input of the indicator recommendation model, and the SVD++ algorithm is used for training to obtain a trained indicator recommendation model.

[0058] According to a third aspect of the present disclosure, there is provided an electronic device, including:

[0059] at least one processor; and

[0060] a memory communicatively connected to the at least one processor; wherein,

[0061] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect.

[0062] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the method described in the first aspect.

[0063] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method as described in the first aspect above.

[0064] The method, device, electronic device and storage medium for index recommendation provided by the present disclosure determine the user identification information and the user's current target operation area, call the target index recommendation model corresponding to the target operation area, and generate the corresponding recommendation index for the user identification information based on the target index recommendation model, wherein different operation areas generate corresponding recommendation indexes through different index recommendation models, and output and display the content corresponding to the recommendation index. Combining multi-dimensional user data (user data corresponding to different operation areas) and business scenario information, more accurate index recommendation is achieved.

[0065] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0067] Figure 1 A flowchart of a method for recommending indicators provided by an embodiment of the present disclosure;

[0068] Figure 2A flowchart of another indicator recommendation method provided by an embodiment of the present disclosure;

[0069] Figure 3 A flowchart of another indicator recommendation method provided by an embodiment of the present disclosure;

[0070] Figure 4 A flowchart of another indicator recommendation method provided by an embodiment of the present disclosure;

[0071] Figure 5 A flowchart of another indicator recommendation method provided by an embodiment of the present disclosure;

[0072] Figure 6 A schematic diagram of the structure of an indicator recommendation device provided in an embodiment of the present disclosure;

[0073] Figure 7 A schematic diagram of the structure of another indicator recommendation device provided in an embodiment of the present disclosure;

[0074] Figure 8 A schematic block diagram of an exemplary electronic device provided for an embodiment of the present disclosure. DETAILED DESCRIPTION

[0075] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0076] The following describes the index recommendation method, device, electronic device and storage medium of the embodiments of the present disclosure with reference to the accompanying drawings.

[0077] Figure 1 A flowchart of a method for recommending indicators provided in an embodiment of the present disclosure.

[0078] like Figure 1 As shown, the method comprises the following steps:

[0079] Step 101, determining user identification information and the user's current target operation area;

[0080] The operation area described in the embodiment of the present application includes a first operation area (corresponding to the guess you like area), a second operation area (corresponding to the guess you want to watch area), a third operation area (corresponding to the watched again area) and a fourth operation area (corresponding to the guess you want to search area).

[0081] In another implementation of the embodiment of the present application, the operation area may include three operation areas or five operation areas, and the corresponding specific module contents are also different. Specifically, the embodiment of the present application does not limit the specific operation areas included in the recommendation system. The corresponding operation areas need to be set in combination with the needs of users in different scenarios.

[0082] Step 102, calling the target indicator recommendation model corresponding to the target operation area, and generating corresponding recommendation indicators for the user identification information based on the target indicator recommendation model, wherein different operation areas generate corresponding recommendation indicators through different indicator recommendation models.

[0083] In an embodiment of the present application, an independent target indicator recommendation model is configured for each operation area. The target indicator recommendation model is obtained by analyzing the behavior data of each user in the corresponding operation area. For example, the target indicator recommendation model corresponding to the area you want to search is obtained by predicting the user's search preference through the user's search behavior, etc.

[0084] It should be noted that when making recommendations to the user, when the target indicator recommendation model is used for recommendation, the target indicator recommendation model may eventually output multiple recommendation indicators, and the optimal indicator (the indicator that the user is most interested in or most wants to click) is selected from these recommendation indicators.

[0085] In another implementation of the embodiment of the present application, when determining the recommendation indicator based on the target indicator recommendation model, multiple recommendation indicators can all be displayed as final recommendation results for the user to select based on the displayed multiple recommendation results.

[0086] It should be noted that the target indicator recommendation model described in the embodiment of the present application is a real-time model, which analyzes the recommended targets that the user may be interested in or click through user information in the corresponding operation area.

[0087] Step 103: Output and display the content corresponding to the recommendation indicator.

[0088] The method, device, electronic device and storage medium for index recommendation provided by the present disclosure determine the user identification information and the user's current target operation area, call the target index recommendation model corresponding to the target operation area, and generate the corresponding recommendation index for the user identification information based on the target index recommendation model, wherein different operation areas generate corresponding recommendation indexes through different index recommendation models, and output and display the content corresponding to the recommendation index. Combining multi-dimensional user data (user data corresponding to different operation areas) and business scenario information, more accurate index recommendation is achieved.

[0089] The embodiment of the present application is explained with four different indicator recommendation models, and the principles of the construction method and the method in the application stage are the same. The difference is that the data used in the training of the four indicator recommendation models are different, and the purpose is to be closer to the real needs of users.

[0090] Figure 2 As shown, Figure 2 Another indicator recommendation method provided in the embodiment of the present application includes:

[0091] Step 201 , parse the user operation history data to obtain the operation history log and the respectively associated user basic information and indicator basic information.

[0092] Through in-depth analysis of user operation history and behavior data, we get the operation history log, and use the operation history log as the main table to associate the user basic information table and the indicator basic information table to obtain user characteristics and indicator characteristics respectively. The input feature data is as follows:

[0093]

[0094] Step 202: construct a dynamic factor matrix according to the operation history log, the user basic information and the indicator basic information.

[0095] When constructing a dynamic factor matrix, the following methods can be used but are not limited to: Figure 3 Shown, including

[0096] Step 301, calculating the degree of attention of each user to the indicator according to the operation history log, the user basic information and the indicator basic information.

[0097] First, based on the user basic information and the indicator basic information, calculate the user's score for a single indicator and the average of the user's scores for all indicators, and perform cosine similarity calculation based on the user's score for a single indicator and the average of the user's scores for all indicators to obtain the degree of attention of each user to the indicator.

[0098] In the recommendation system, although the traditional collaborative filtering method based on cosine similarity can capture the similarity between users and indicators to a certain extent, in actual business scenarios, its accuracy and effectiveness are often affected by the user's scoring scale, indicator diversity and user behavior complexity. Therefore, the embodiment of the present application proposes a precise indicator recommendation method based on multi-dimensional collaborative filtering and behavior analysis, which aims to achieve more accurate indicator recommendation by comprehensively considering user behavior, indicator attributes and multi-dimensional similarity on the basis of the traditional cosine similarity system filtering method.

[0099] Since the cosine similarity method in the prior art does not take into account the issue of user scoring scale, for example, some people like to give high scores while others like to give low scores. Due to the different scoring habits, the calculated similarity may be very different from the actual one. Therefore, in order to solve this problem, the embodiment of the present application proposes a modified cosine similarity calculation. The specific formula is as follows:

[0100]

[0101] Among them, R u,i and R u,j represents the score of user u on indicators i and j, R u It represents the average score of all indicators given by user u.

[0102] Step 302: construct the dynamic factor matrix according to each user's attention level to the indicator.

[0103] Define a dynamic factor matrix U, which not only contains the static attention relationship between users and indicators, but also incorporates dynamic factors such as time and user behavior changes. By updating the dynamic factor matrix in real time, it can more accurately reflect the user's attention to indicators at different time points and in different scenarios.

[0104] The construction of the dynamic factor matrix is ​​based on a huge amount of user operation history and behavior data. Each element Umn represents the degree of attention of user m to indicator n. This degree can be calculated by the calculation method of 301 above, or by complex algorithms and models. Specifically, this embodiment of the present application is not limited.

[0105]

[0106] Step 303: decompose the dynamic factor matrix into the product of the user and hidden trait matrix, and the hidden trait and indicator matrix.

[0107] Decompose the dynamic factor matrix U into the product of two sub-matrices. One is the user-trait matrix u mk , and the other is the trait-indicator matrix i kn .

[0108]

[0109] Among them, u mk It can be regarded as the user m’s preference for the hidden characteristic k of the indicator, and i kn It can be regarded as the degree to which trait k is reflected in indicator n.

[0110] User-feature matrix u mk Represents a user's preference, interest or need, the trait-indicator matrix ikn It represents the degree to which trait k is reflected in indicator n.

[0111] Step 203: Perform dimensionality reduction processing on the dynamic factor matrix to obtain a local factor matrix.

[0112] When calculating the local factor matrix, the following methods can be used but are not limited to: Figure 4 As shown, including:

[0113] Step 401 , performing feature fusion processing on the user and hidden trait matrix, and the hidden trait and indicator matrix to obtain a fused matrix.

[0114] In actual business, the interaction data between users and indicators constitutes the core data source of the recommendation system. When a user browses an indicator, this behavior not only indicates that the user has been exposed to the indicator, but also implies that the user has a certain preference or interest in the indicator. This implicit user feedback is extremely valuable for the recommendation system because it can reveal the user's true preferences and needs.

[0115] In order to deeply capture this implicit feedback and break through the limitations of traditional recommendation systems, the embodiment of the present application proposes a deep fusion technology of enhanced factor matrix decomposition and user behavior based on SVD++. This technology not only considers the user's explicit feedback, but also integrates implicit feedback such as the user's browsing behavior into the model, thereby more accurately characterizing the user's personalized preferences.

[0116] Specifically, when a user browses an indicator, it can be regarded as an implicit feedback, and this feedback can be reflected in the integration model by introducing implicit parameters. This processing method enables the integration model - SVD++ enhanced version, to capture the user's true preferences more finely.

[0117] Step 402: Call the FunkSVD algorithm to optimize the fused matrix to obtain a local factor matrix.

[0118] First, the global factor matrix is ​​reduced to a local factor matrix, and then the FunkSVD algorithm is used for further optimization. By only decomposing N-1 matrices instead of the traditional N matrices, the computational complexity can be significantly reduced and the sparsity problem can be alleviated.

[0119] The following is the specific matrix optimization process:

[0120] Optimize the matrix M:

[0121]

[0122] The FunkSVD algorithm is used for further optimization to minimize the score residuals of the user's indicators. In other words, the mean square error can be used as the loss function to find the final P and Q.

[0123] For user index m ij , if the algorithm is used to decompose the matrix, the corresponding representation is Using the mean square error loss function, we expect As small as possible, if all combinations of indicators and samples are considered, the minimization formula is as follows:

[0124]

[0125] In the indicator recommendation process, in order to prevent overfitting, an L2 regularization term is added, so the objective function after optimization is:

[0126]

[0127] Where λ is the regularization coefficient, which needs to be adjusted. For this optimization problem, the result is generally obtained by optimizing through the gradient descent method. i ,q j Derivatives respectively:

[0128]

[0129] When using the gradient descent method iteratively, p i ,q j The iteration formula is:

[0130]

[0131] Finally, the local factor matrix is ​​decomposed to the optimal result.

[0132] The embodiment of the present application combines user behavior data with factor matrix decomposition. By conducting an in-depth analysis of the collection of user browsing indicators, the user's hidden preference bias is extracted, and the recommendation results are optimized. This deep fusion method not only reduces the number of factors, but also improves the accuracy and efficiency of recommendations.

[0133] The efficient collaborative filtering algorithm based on local factor matrix optimization combines the FunkSVD algorithm with local factor matrix optimization, overcoming the shortcomings of the traditional algorithm by reducing the number of matrix decompositions and introducing regularization terms. This method not only improves the accuracy of recommendations, but also significantly reduces the computational cost, providing an efficient and effective solution for indicator recommendations in actual business scenarios.

[0134] Step 204, constructing different indicator recommendation models according to the local factor matrix, wherein the indicator recommendation model includes a first indicator recommendation model, a second indicator recommendation model, a third indicator recommendation model and a fourth indicator recommendation model;

[0135] When constructing an indicator recommendation model, the following methods can be used but are not limited to: Figure 5 As shown, including:

[0136] Step 501, constructing the first indicator recommendation model based on the similarity between users in the operation history log;

[0137] The first indicator recommendation model corresponds to the "Guess you like" operation area, which is located in the "All" page of the indicator hall and the "Recommendation" page column. This recommendation column has no clear purpose and should consider user preferences for recommendations. Therefore, the "Guess you like" recommendation model (first indicator recommendation model) is based on the operation log data of the operation types of "large card browsing", "insight" and "attention" in the historical data as positive samples to predict the probability distribution of user click preference indicators.

[0138] Result output:

[0139]

[0140] Step 502: construct the second indicator recommendation model based on the operation log data of the operation type of concern in the operation history log;

[0141] The second indicator recommendation model corresponds to the guess what you want to see operation area, which is located in the relevant recommendation column of the "Follow" page of the indicator hall. This recommendation column is highly indicative and should be recommended from perspectives such as strong correlation indicators and related user attention indicators. Therefore, the second indicator recommendation model is based on the operation log data of the operation type "Follow" in the historical data as a positive sample to predict the probability distribution of user searches.

[0142] Result output:

[0143]

[0144] Step 503: construct the third indicator recommendation model based on the operation log data in the operation history log whose operation type is search.

[0145] The third indicator recommendation model corresponds to the "Guess what you want to search" operation area, which is located in the "Guess what you want to search" column on the indicator hall search page. This recommendation column has a clear purpose, and the user's search target may be a known indicator or an indicator of interest. Therefore, the third indicator recommendation model is based on the operation log data of the operation type "search" in the historical data as a positive sample to predict the probability distribution of user searches.

[0146] Result output:

[0147]

[0148] Step 504: Based on the indicator information of the current operation type in the operation history log, the fourth indicator recommendation model is constructed according to the indicator change trend, correlation analysis, and influencing factor identification.

[0149] When users operate a certain indicator, they may need to view or operate similar indicators due to business needs. It is necessary to recommend similar indicators based on the currently operated indicator, also known as "looked again" indicator recommendation. By combining text data with other indicator information for similarity calculation, other information includes indicator type, indicator source, indicator launch time, etc., similar indicators of the "looked again" module can be predicted and recommended.

[0150] Result output:

[0151]

[0152] To sum up, the precise indicator recommendation method based on multi-dimensional collaborative filtering and behavioral analysis proposed in the embodiment of the present application can more accurately capture the relationship between users and indicators, achieve more precise indicator recommendations, and solve the shortcomings of traditional collaborative filtering methods in actual business scenarios. It not only improves the accuracy of recommendations, but also enhances user experience.

[0153] Step 205 , training different indicator recommendation models respectively to obtain trained indicator recommendation models.

[0154] The local factor matrix is ​​used as the input of the indicator recommendation model, and the SVD++ algorithm is used for training to obtain a trained indicator recommendation model.

[0155] The training of the first indicator recommendation model includes:

[0156] a. User portrait construction: Collect and analyze users' historical behavior data, including visit records, click behaviors, dwell time, etc., and combine the user's basic information (such as department, position, etc.) to construct a multi-dimensional user portrait.

[0157] b. Index recommendation generation: Use collaborative filtering algorithms to calculate the similarity between users. Based on similarity sorting, find other user groups with similar interests to the user. At the same time, based on the index access records of similar users, combined with the current user's business background and needs, screen out the most likely indicators of interest and recommend them to the user in a certain sorting method.

[0158] Using big data processing frameworks such as Hadoop or Spark, the user behavior data is cleaned, integrated and preprocessed. The similarity between users is calculated using methods such as cosine similarity and Pearson correlation coefficient. Finally, the recommendation results are optimized and personalized in combination with the machine learning model to obtain the first indicator recommendation model after training.

[0159] The training of the second indicator recommendation model includes:

[0160] For users with clear business objectives and existing indicator-focused behaviors, the recommendation mechanism can derive other similar indicators by analyzing the similarity of the indicators that users have already focused on, and display them in the "Guess what you want to see" operation area (module);

[0161] a. Analysis of indicators that users have paid attention to: Analyze the indicator data that users have paid attention to, including the visit frequency, residence time, correlation analysis, etc. of the indicators.

[0162] Similar indicator mining and generation: Using the correlation and similarity between indicators, we can find other indicators that are similar or related to the indicators we have focused on through clustering, association rule mining and other methods. Based on the results of similar indicator mining, we can use collaborative filtering algorithms to generate a recommendation list and show it to users.

[0163] During training, a graph database (such as Neo4j) or a relational database is used to build an association network between indicators, and association rule mining algorithms such as clustering algorithms are applied to find the association between indicators. Finally, combined with user preference settings and real-time behavior data, a collaborative filtering algorithm is used to generate recommendation results and personalize and optimize the recommendation results to obtain a trained second indicator recommendation model.

[0164] The training of the third indicator recommendation model includes:

[0165] For users with clear index requirements, the recommendation mechanism can combine the user's historical access data, seasonal characteristics, business characteristics and other factors to calculate the index that the user most wants to search and view at the moment, and display it in the "Guess what you want to search" module;

[0166] a. Search behavior analysis: real-time collection and analysis of users’ search records, including search keywords, search time, search result clicks, etc.

[0167] b. Keyword extraction and matching: Use natural language processing (NLP) technology to perform operations such as word segmentation and stop word removal on the user's search content to extract keywords. Use the keyword matching algorithm to find the indicator that best matches the user's search intent.

[0168] c. Search recommendation generation: Combine the user's search history, business characteristics, current season and other factors to predict the user's current search needs, and use collaborative filtering algorithms to generate search index recommendations.

[0169] During training, Elasticsearch search engine technology is used to achieve efficient and real-time search functions. At the same time, NLP algorithms such as TF-IDF and Word2Vec are applied for keyword extraction and semantic analysis. Finally, the collaborative filtering algorithm is used, combined with business rules and machine learning models, to dynamically adjust and optimize the search recommendation results to obtain the trained third indicator recommendation model.

[0170] The training of the fourth indicator recommendation model includes:

[0171] For users with clear business objectives who want to analyze the key factors that affect business performance indicators, the recommendation mechanism can combine the correlation of indicators and display multiple key indicators that affect business performance indicators in the "Viewed Again" module.

[0172] a. Key indicator analysis: In-depth analysis of key factors and indicators that affect business performance, including indicator change trends, correlation analysis, and identification of influencing factors.

[0173] b. Generate relevant indicator recommendations: Based on the analysis of key business indicators and combined with the user's business background and needs, collaborative filtering algorithms are used to recommend other relevant indicators that have an important impact on business decisions.

[0174] During training, data mining and machine learning techniques are used to conduct in-depth analysis and mining of a large amount of indicator data. Decision tree, random forest and other algorithms are used to build a business performance prediction model to identify key indicators and influencing factors. Finally, collaborative filtering algorithm and random forest regression are used to build a model to generate other indicator recommendations related to key indicators, and the fourth indicator recommendation model after training is obtained.

[0175] In summary, the embodiment of the present application combines factor matrix enhancement technology with multi-dimensional user data and business scenario information to achieve personalized indicator display and recommendation of data services that vary from person to person and from position to position. This method not only improves the accuracy and efficiency of the recommendation system, but also enables dynamic optimization and adjustment according to changes in business scenarios and user needs, providing users with a better data analysis experience.

[0176] The embodiments of the present application can achieve the following beneficial effects:

[0177] 1. Decomposition of the global factor matrix

[0178] In order to solve the problem that the global factor matrix has a large amount of calculation and is easy to overfit the explicit attribute indicators, the embodiment of the present application adopts the SVD++ enhanced version method for matrix decomposition. This method not only considers the user's display feedback, but also incorporates the user's browsing behavior as implicit feedback into the model to decompose the factor matrix. After reducing the number of factors, the gradient calculation method is introduced to minimize the objective function, thereby accurately capturing the user's personalized preferences. This method not only considers the implicit layer relationship of the user's preference for indicators, deeply integrates the user's behavioral characteristics, but also can more quickly decompose the global factor matrix into a local factor matrix, improves the calculation efficiency and reduces the risk of overfitting.

[0179] 2. Personalized optimization and calculation of local factor matrix:

[0180] After obtaining the local factor matrix, the embodiment of the present application further uses the FunkSVD method to optimize these matrices. Different from the traditional FunkSVD algorithm that uses the idea of ​​linear regression, the embodiment of the present application combines the FunkSVD algorithm with the optimization of the local factor matrix, focusing on minimizing the score residuals between users and indicators, reducing the number of decomposed matrices, and preventing overfitting by introducing the L2 regularization term. The gradient descent method is used for iterative solution to ensure that each local factor matrix can accurately reflect the user's indicator preferences in a specific scenario, thereby improving the accuracy of recommendations and reducing computational costs. The optimized local factor matrix can more accurately recommend indicators that best meet the needs of users through the calculation of cosine similarity based on indicators, combined with user preferences and behavioral habits, thereby improving the accuracy and personalization of recommendations.

[0181] 3. Construction and application of local factor matrix based on specific indicator recommendation scenarios:

[0182] In the process of constructing a local factor matrix, the embodiment of the present application innovatively combines factor matrix enhancement technology with multi-dimensional user data and business scenarios to achieve personalized recommendations. In each scenario, by deeply analyzing user behavior and business needs, mining data such as user portraits, search behaviors, attention indicators, and business key indicators, and using clustering, association rule mining, natural language processing and other technologies to extract key information. Then, combined with the collaborative filtering algorithm, the local factor matrix is ​​optimized to generate an accurate indicator recommendation list. This method fully considers the characteristics of specific indicator recommendation scenarios and user behavior patterns. By deeply mining and analyzing user historical operation data, basic features, and basic features of indicators, the user's indicator attention level and preference characteristics in specific scenarios are accurately portrayed, and four indicator recommendation scenarios of "Guess you like, guess you want to watch, watch again, and guess you want to search" are constructed.

[0183] Corresponding to the above-mentioned index recommendation method, the present invention also proposes an index recommendation device. Since the device embodiment of the present invention corresponds to the above-mentioned method embodiment, details not disclosed in the device embodiment can be referred to the above-mentioned method embodiment, and will not be repeated in the present invention.

[0184] Figure 6 A schematic diagram of a device for recommending indicators provided by an embodiment of the present disclosure is shown in FIG. Figure 6 As shown, including:

[0185] A determination unit 61, used to determine user identification information and a user's current target operation area;

[0186] A first processing unit 62 is used to call a target indicator recommendation model corresponding to the target operation area, and generate a corresponding recommendation indicator for the user identification information based on the target indicator recommendation model, wherein different operation areas generate corresponding recommendation indicators through different indicator recommendation models;

[0187] The display unit 63 is used to output and display the content corresponding to the recommendation indicator.

[0188] The indicator recommendation device provided by the present disclosure determines the user identification information and the user's current target operation area, calls the target indicator recommendation model corresponding to the target operation area, and generates the corresponding recommendation indicator for the user identification information based on the target indicator recommendation model, wherein different operation areas generate corresponding recommendation indicators through different indicator recommendation models, and outputs and displays the content corresponding to the recommendation indicator. Combining multi-dimensional user data (user data corresponding to different operation areas) and business scenario information, more accurate indicator recommendation is achieved.

[0189] Furthermore, in a possible implementation of the embodiment of the present disclosure, as Figure 7 As shown, the device also includes:

[0190] The parsing unit 64 is used to parse the user operation history data to obtain the operation history log and the respectively associated user basic information and indicator basic information before calling the target indicator recommendation model corresponding to the target operation area and generating the corresponding recommendation indicator for the user identification information based on the target indicator recommendation model;

[0191] A first construction unit 65 is used to construct a dynamic factor matrix according to the operation history log, the user basic information and the indicator basic information;

[0192] A second processing unit 66 is used to perform dimensionality reduction processing on the dynamic factor matrix to obtain a local factor matrix;

[0193] A second construction unit 67 is used to construct different indicator recommendation models according to the local factor matrix, wherein the indicator recommendation model includes a first indicator recommendation model, a second indicator recommendation model, a third indicator recommendation model and a fourth indicator recommendation model;

[0194] The training unit 68 is used to train different indicator recommendation models respectively to obtain trained indicator recommendation models.

[0195] Furthermore, in a possible implementation of the embodiment of the present disclosure, as Figure 7 As shown, the first construction unit 65 includes:

[0196] A calculation module 651 is used to calculate the degree of attention of each user to the indicator according to the operation history log, the user basic information and the indicator basic information;

[0197] A construction module 652 is used to construct the dynamic factor matrix according to the degree of attention paid by each user to the indicator;

[0198] The decomposition module 653 is used to decompose the dynamic factor matrix into the product of the user and hidden trait matrix and the hidden trait and indicator matrix.

[0199] Furthermore, in a possible implementation of the embodiment of the present disclosure, as Figure 7 As shown, the calculation module 651 is also used for:

[0200] Calculate the user's score for a single indicator and the average of the user's scores for all indicators based on the user's basic information and the indicator basic information;

[0201] The cosine similarity calculation is performed based on the user's score for a single indicator and the average value of the user's score for all indicators to obtain the degree of attention of each user to the indicator.

[0202] Furthermore, in a possible implementation of the embodiment of the present disclosure, as Figure 7 As shown, the building module 652 is also used for:

[0203] Perform feature fusion processing on the user and hidden trait matrix, and the hidden trait and indicator matrix to obtain a fused matrix;

[0204] The FunkSVD algorithm is called to optimize the fused matrix to obtain a local factor matrix.

[0205] Furthermore, in a possible implementation of the embodiment of the present disclosure, as Figure 7 As shown, the second construction unit 67 is also used for:

[0206] Based on the similarities between users in the operation history log, construct the first indicator recommendation model;

[0207] Building the second indicator recommendation model based on the operation log data of the operation type of concern in the operation history log;

[0208] Based on the operation log data in the operation history log whose operation type is search, construct the third indicator recommendation model;

[0209] Based on the indicator information of the operation type of the current operation in the operation history log, the fourth indicator recommendation model is constructed according to the indicator change trend, correlation analysis, and influencing factor identification.

[0210] Furthermore, in a possible implementation of the embodiment of the present disclosure, as Figure 7 As shown, the training unit is also used for:

[0211] The local factor matrix is ​​used as the input of the indicator recommendation model, and the SVD++ algorithm is used for training to obtain a trained indicator recommendation model.

[0212] It should be noted that the above explanation of the method embodiment is also applicable to the device of the embodiment of the present disclosure, and the principle is the same, which is no longer limited in the embodiment of the present disclosure.

[0213] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0214] Figure 8 A schematic block diagram of an example electronic device 800 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0215] like Figure 8As shown, the device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a ROM (Read-Only Memory) 802 or a computer program loaded from a storage unit 808 to a RAM (Random Access Memory) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An I / O (Input / Output) interface 805 is also connected to the bus 804.

[0216] A number of components in the device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0217] The computing unit 801 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a CPU (Central Processing Unit), a GPU (Graphic Processing Units), various dedicated AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, a DSP (Digital Signal Processor), and any appropriate processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as the method for index recommendation. For example, in some embodiments, the method for index recommendation may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the method described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to execute the aforementioned indicator recommendation method in any other appropriate manner (for example, by means of firmware).

[0218] Various embodiments of the systems and techniques described above herein may be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application Specific Standard Products), SOCs (System On Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include: being implemented in one or more computer programs that may be executed and / or interpreted on a programmable system including at least one programmable processor that may be a special purpose or general purpose programmable processor that may receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0219] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0220] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a RAM, a ROM, an EPROM (Electrically Programmable Read-Only-Memory) or a flash memory, an optical fiber, a CD-ROM (Compact Dis sc Read-Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0221] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball), through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0222] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: LAN (Local Area Network), WAN (Wide Area Network), the Internet, and blockchain networks.

[0223] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0224] It should be noted that artificial intelligence is a discipline that studies how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), and includes both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning / deep learning, big data processing technology, knowledge graph technology, and other major directions.

[0225] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0226] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for index recommendation, characterized in that: include: Determine user identification information and the user's current target operation area; Calling a target indicator recommendation model corresponding to the target operation area, and generating a corresponding recommendation indicator for the user identification information based on the target indicator recommendation model, wherein different operation areas generate corresponding recommendation indicators through different indicator recommendation models; The content corresponding to the recommendation indicator is output and displayed.

2. The method according to claim 1, characterized in that Before calling the target indicator recommendation model corresponding to the target operation area and generating the corresponding recommendation indicator for the user identification information based on the target indicator recommendation model, the method further includes: Analyze the user operation history data to obtain the operation history log and the associated user basic information and indicator basic information; Constructing a dynamic factor matrix according to the operation history log, the user basic information and the indicator basic information; Performing dimensionality reduction processing on the dynamic factor matrix to obtain a local factor matrix; Constructing different indicator recommendation models according to the local factor matrix, wherein the indicator recommendation model includes a first indicator recommendation model, a second indicator recommendation model, a third indicator recommendation model and a fourth indicator recommendation model; Different indicator recommendation models are trained respectively to obtain trained indicator recommendation models.

3. The method according to claim 2, characterized in that The constructing of a dynamic factor matrix according to the operation history log, the user basic information and the indicator basic information includes: Calculate the attention level of each user to the indicator according to the operation history log, the user basic information and the indicator basic information; Constructing the dynamic factor matrix according to each user's attention to the indicator; The dynamic factor matrix is ​​decomposed into the product of the user and hidden trait matrix, and the hidden trait and indicator matrix.

4. The method according to claim 3, characterized in that The calculating, according to the operation history log, the user basic information and the indicator basic information, respectively, the degree of attention of each user to the indicator comprises: Calculate the user's score for a single indicator and the average of the user's scores for all indicators based on the user basic information and the indicator basic information; The cosine similarity calculation is performed based on the user's score for a single indicator and the average value of the user's score for all indicators to obtain the degree of attention of each user to the indicator.

5. The method according to claim 3, characterized in that: The dimensionality reduction process of the dynamic factor matrix to obtain a local factor matrix comprises: Performing feature fusion processing on the user and hidden trait matrix, and the hidden trait and indicator matrix to obtain a fused matrix; The FunkSVD algorithm is called to optimize the fused matrix to obtain a local factor matrix.

6. The method according to claim 5, characterized in that The constructing different indicator recommendation models according to the local factor matrix includes: Based on the similarities between users in the operation history log, construct the first indicator recommendation model; Building the second indicator recommendation model based on the operation log data of the operation type of concern in the operation history log; Based on the operation log data in the operation history log whose operation type is search, construct the third indicator recommendation model; Based on the indicator information of the operation type of the current operation in the operation history log, the fourth indicator recommendation model is constructed according to the indicator change trend, correlation analysis, and influencing factor identification.

7. The method according to claim 6, characterized in that The training of different indicator recommendation models respectively to obtain the trained indicator recommendation models includes: The local factor matrix is ​​used as the input of the indicator recommendation model, and the SVD++ algorithm is used for training to obtain a trained indicator recommendation model.

8. A device for index recommendation, characterized in that: include: A determination unit, used to determine user identification information and the user's current target operation area; A first processing unit is used to call a target indicator recommendation model corresponding to the target operation area, and generate a corresponding recommendation indicator for the user identification information based on the target indicator recommendation model, wherein different operation areas generate corresponding recommendation indicators through different indicator recommendation models; The display unit is used to output and display the content corresponding to the recommendation indicator.

9. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-7.

Images