Data asset recommendation method, device, computer equipment and storage medium

Abstract

The embodiments of the present application provide a data asset recommendation method, device, computer equipment and storage medium, which can be applied to the field of artificial intelligence technology. The method includes: obtaining a data search keyword for a first user; performing feature engineering processing on the user data of the first user according to a user feature set to obtain user feature data; performing feature engineering processing on M data assets according to a data asset feature set to obtain M data asset feature data; using a data asset recommendation model to process the data search keyword, user feature data, and M data asset feature data to obtain data asset recommendation probabilities corresponding to N data assets; determining at least one data asset to be pushed to the first user according to the data asset recommendation probabilities corresponding to the N data assets. Implementing the embodiments of the present application helps to improve the accuracy of data asset recommendations.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and specifically to a data asset recommendation method, device, computer equipment and storage medium. Background Art

[0002] As the number of data assets continues to increase, the time required to manually find a specified data asset is also increasing. To solve this problem, a data asset recommendation system has emerged. In a data asset recommendation system, the system needs to match based on the search information entered by the user and push the successfully matched data assets to the user.

[0003] The inventors realized that the way in which the data asset recommendation system recommends data assets to users is still at the most primitive metadata management level, that is, the user is asked to enter the library name, the English name of the table, and the English name of the field, and then perform an accurate search to push the matching data assets to the user. However, users often do not know the precise information of the data assets. For example, users usually only know the field information to which the data assets to be queried belong. At this time, it is impossible to push the required data assets to the user, or the pushed data assets are inaccurate and have low reliability. Summary of the invention

[0004] The embodiments of the present application provide a data asset recommendation method, apparatus, computer device, and storage medium, which help to improve the accuracy of data asset recommendations.

[0005] The first aspect of the present application provides a data asset recommendation method, comprising:

[0006] receiving data search information input by a first user, identifying the data search information, and obtaining a data search keyword;

[0007] Acquire user data of the first user, and perform feature engineering processing on the user data of the first user according to the user feature set to obtain user feature data;

[0008] Obtain M data assets, and perform feature engineering on the M data assets according to the data asset feature set to obtain M data asset feature data, where M is a positive integer;

[0009] The data asset recommendation model is used to process the data search keywords, user feature data, and M data asset feature data to obtain the data asset recommendation probabilities corresponding to the N data assets, where N is a positive integer less than or equal to M;

[0010] At least one data asset to be pushed to the first user is determined according to the data asset recommendation probabilities respectively corresponding to the N data assets.

[0011] A second aspect of the present application provides a data asset recommendation device, comprising:

[0012] A receiving module, configured to receive data search information input by a first user, identify the data search information, and obtain a data search keyword;

[0013] an acquisition module, used to acquire user data of a first user, and perform feature engineering processing on the user data of the first user according to a user feature set to obtain user feature data; used to acquire M data assets, and perform feature engineering processing on the M data assets respectively according to the data asset feature set to obtain M data asset feature data, where M is a positive integer;

[0014] A processing module, used to process the data search keyword, the user characteristic data, and the M data asset characteristic data using the data asset recommendation model to obtain the data asset recommendation probabilities corresponding to the N data assets, where N is a positive integer less than or equal to M;

[0015] The data asset recommendation module is used to determine at least one data asset to be pushed to the first user according to the data asset recommendation probabilities respectively corresponding to the N data assets.

[0016] The third aspect of the present application provides a computer device, including a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute the method of the first aspect of the embodiment of the present application.

[0017] The fourth aspect of the present application provides a computer-readable storage medium, which stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, the processor executes the method of the first aspect of the embodiment of the present application.

[0018] This application obtains a data search keyword by identifying the data search information input by the first user, and then performs feature engineering processing on the user data of the first user obtained based on the user feature set to obtain user feature data, and performs feature engineering processing on M data assets based on the data asset feature set to obtain M data asset feature data; then the data asset recommendation model is used to process the data search keyword, user feature data, and N data asset feature data to obtain the data asset recommendation probabilities corresponding to the N data assets, and then at least one data asset to be pushed to the first user can be determined based on the data asset recommendation probabilities corresponding to the N data assets. The implementation of the embodiment of the present application enables data asset recommendation to be implemented based on the data search information input by the user, user feature data, data asset feature data, and the data asset recommendation model, which helps to improve the accuracy of data asset recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 This is a schematic diagram of a data asset recommendation system architecture provided by an embodiment of the present application;

[0021] Figure 2 It is a flowchart of a data asset recommendation method provided in an embodiment of the present application;

[0022] Figure 3 It is a structural diagram of a data asset recommendation device provided in an embodiment of the present application;

[0023] Figure 4 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The following will describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0025] The data asset recommendation method of the present application can be applied to a computer device to implement the recommendation of data assets. Optionally, the computer device can be a terminal device, a server, a data platform or other data asset recommendation device.

[0026] For example, see Figure 1 , Figure 1 This is a schematic diagram of the architecture of a data asset recommendation system provided by this application. Figure 1 As shown, the system includes a server 100 and multiple terminal devices (or clients), for example, the multiple terminal devices may include terminal device 101, terminal device 102, and terminal device 103. Among them, terminal device 101, terminal device 102, and terminal device 103 can communicate with each other through the network with server 100. For example, the terminal device can receive data search information input by the user and send the data search information to the server. The server can execute the data asset recommendation method of the present application and push one or more data assets to the user of the terminal device. For another example, the server may store user data, data assets and other information, and the terminal device can obtain this information from the server to implement the data asset recommendation method of the present application, and so on.

[0027] Optionally, the terminal device may be a mobile phone, tablet computer, laptop computer, PDA, mobile internet device (MID), wearable device (such as smart watch, smart bracelet, etc.) or other terminal. The server may be an independent server, or a cloud server that provides basic cloud computing services such as cloud services, cloud database, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery network (CDN), and big data and artificial intelligence platforms, etc., which is not limited in this application.

[0028] The technical solution of the present application can be applied to the field of artificial intelligence and / or big data technology, wherein artificial intelligence (AI) is a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. Based on artificial intelligence, it is possible to accurately push data assets to users based on data input by users, thereby promoting the construction of smart cities. Optionally, the data involved in this application may be stored in a database, a cloud platform, a blockchain node, etc., or may also be stored in a terminal device, which is not limited in this application. Further optionally, the terminal device and / or server may be a node in a blockchain.

[0029] The present application can obtain search keywords by identifying data search information, and push at least one data asset to the user based on the search keywords, user feature data, asset feature data, etc., so that the data assets can be pushed based on the data search information input by the user to improve the accuracy of data asset recommendations.

[0030] Optionally, the technical solution of the present application can be applied to various data asset push scenarios, such as medical data asset push scenarios in digital medicine, and financial data asset push scenarios in financial technology. For example, medical data assets can be data assets corresponding to various medical data. Optionally, the medical data can include one or more of personal health records, prescriptions, examination reports, and medical insurance business data. Further optionally, medical data assets and / or financial data assets can be stored in a blockchain.

[0031] The embodiments of the present application provide a data asset recommendation method, device, computer equipment and medium for improving the reliability of data asset recommendation. Among them, the user data obtained by the present application is authorized by the user. The following are detailed descriptions.

[0032] See also Figure 2 , Figure 2 is a flow chart of a data asset recommendation method provided in an embodiment of the present application. Figure 2 As shown, the data asset recommendation method may include the following steps 201-204:

[0033] 201. Receive data search information input by a first user, identify the data search information, and obtain a data search keyword.

[0034] The data search information may be used to describe the information of the required data assets for searching the data assets. For example, the data search information may include domain information or other characteristic information of the data assets, so as to determine the data assets required by the user according to the keywords of the data search information, i.e., the data search keywords.

[0035] In some embodiments, the data search information input by the first user may be received by reading the data typed into the input box by the user to obtain the data search information. The input box is set by the terminal device, and the input box may be a voice input box or a text input box, which is not limited here.

[0036] In some embodiments, the method for receiving the data search information may also be: determining the data search information from the recommendation bar set in the terminal device based on the user's click instruction, specifically, determining the recommended information from the recommendation bar according to the click instruction, and determining the recommended information as the data search information. The recommendation bar contains at least one recommendation information, such as the recommendation information may be a domain label or other feature information of the data asset. Optionally, at least one data asset may be sorted from high to low according to the access popularity, and the names corresponding to the first J data assets may be used as the recommendation information, where J is a positive integer, and J may be set by empirical values ​​or historical data, which is not limited here. The access popularity may be the total number of clicks on the data asset by users per unit time, and the unit time may be one hour, one day, or other, which is not limited here.

[0037] Optionally, the data search keyword may be a segmented word in the data search information, or a keyword associated with a segmented word in the data search information, or a segmented word in the data search information and the associated keyword, etc., which are not listed here one by one.

[0038] Among them, a possible method for identifying data search information and obtaining data search keywords may be: deleting interjections from the data search information to obtain first processed data, where interjections may be "ah", "um", etc.; segmenting the first processed data to obtain A second processed data, where A is a positive integer, where the segmentation method may be segmentation based on conjunctions; deleting meaningless words from the A second processed data to obtain B third processed data, where the third processed data may be used as data search keywords, where meaningless words may be conjunctions, prepositions, etc., which are not limited here. Among them, A and B are positive integers, and the size of A and B may be determined based on actual conditions, which are not limited here. For example: assuming that the data search information is "data assets are of medical type and are smart medical", the first processed data is "data assets belong to medical type and are smart medical", the second processed data is "data assets", "belong to medical type", "and are smart medical", and the third processed data is "data assets", "medical type", "smart medical".

[0039] Another possible method for identifying data search information and obtaining data search keywords may be: segmenting the data search information to obtain C words; extracting from the C words based on parts of speech to obtain D words, where D is a positive integer; eliminating meaningless words from the D words to obtain E keywords, where E is a positive integer. For example, meaningless words may be interjections, conjunctions, etc.

[0040] The word segmentation processing method can be to match the word segmented sentence with the words in the dictionary. If the match is successful, the matched part is regarded as a word, and finally a word sequence is generated. The matching method can be a positive matching method, a reverse matching method, a maximum matching method, and a minimum matching method, which are not limited here.

[0041] For example, assuming that the data search information is "Well, I need to search for data assets in the investment field, and the data assets belong to medical", then the C words are "well", "I", "need", "search", "investment", "aspect", "of", "data assets", "and", "belong to", and "medical". Delete the interjections, pronouns, verbs, adverbs, auxiliary words, conjunctions, and prepositions in the C words, and extract the remaining words. The deleted words are "well", "I", "need", "search", "of", "and", and "belong to", and the D words are obtained, which are "investment", "aspect", "data assets", and "medical". If words with no practical meaning include "aspect" and "data assets", the E keywords can be "investment" and "medical".

[0042] 202. Obtain user data of a first user, and perform feature engineering processing on the user data of the first user according to a user feature set to obtain user feature data.

[0043] Among them, the method for obtaining the user data of the first user can be that the terminal device extracts the user data stored in the database, or the terminal device extracts the user data in the blockchain and cloud platform to obtain the user data of the first user.

[0044] User data may be the data of the user of the terminal device. User data may include basic information of the user and behavior information of the user. Basic information may include age, occupation, preferences, etc. Behavior information may include information on data search based on data search keywords, or information related to operations such as whether to respond to recommended data assets (such as click, download, etc.). Operation-related information may include information on responded data assets, information on non-responded data assets, operation time such as response time or search time, name corresponding to the data asset, etc.

[0045] The user feature set may include at least one (one or more) user features, and the user features may include features in terms of the user's basic attributes, behavior information, etc. For example, user features may include user attribute features and user behavior features. Among them, user attribute features may include one or more of the features such as preference features and occupational features; user behavior features may include one or more of the features such as user behavior sequence, contextual features, and statistical features. For example, the user behavior sequence may include a positive feedback sequence of the user's response to the data assets recommended by the terminal device and a negative feedback sequence of the user's failure to respond to the data assets recommended by the terminal device. The response may be operations such as browsing and downloading; contextual features may include one or more of the features such as keywords, the time of the current data request, the operation time of browsing, downloading, etc. in response to the asset data, and the keywords here may be data search keywords for historical data search information, such as the search keywords of the previous few times; statistical features may include information such as field distribution obtained by statistically analyzing the fields to which the data assets to which the user responded belong, such as the subject distribution and library distribution of the user's historical access to assets.

[0046] In a possible implementation, feature engineering is performed on the user data of the first user according to the user feature set, and a method for obtaining the user feature data may be: matching the user data based on at least one user feature in the user feature set, and the matching method may be performing synonym matching on the character string corresponding to at least one user feature in the user feature set and the character string in the user data, respectively. If the match is successful, the position of the user feature in the user data can be determined, and a feature description corresponding to the user feature can be obtained based on the position, and the at least one user feature and the feature description corresponding to the user feature are determined as the user feature data; if the match fails, the feature description corresponding to the user feature is empty.

[0047] For example: the user feature set includes preference features, keywords in context features, and user behavior sequences; the user data is "the user likes sports, and the data search keywords include investment and medical treatment", which is empty and represented by "none". Among them, the preference features in the user feature set match the preferences in the user data, and the keywords in the context features match the data search keywords, and the obtained feature descriptions are "sports", "investment, medical treatment", so the user feature data can be "preference features: sports, keywords in context features: investment, medical treatment, user behavior sequence: none".

[0048] Optionally, the user feature set may be determined based on one or more of the domain features of the data asset, the type of the data asset recommendation system, the identifier of the data asset recommendation system, and the like. For example, different data asset domains may correspond to different user feature sets, different data asset recommendation system types may correspond to different user feature sets, and so on. For example, the types may include financial types, medical types, and so on. This helps to improve the efficiency of feature engineering processing, thereby improving the efficiency of recommendations and improving the reliability of recommendations.

[0049] Before obtaining user data, user authorization is required, or the obtained user data is data authorized by the user. For example, before obtaining the user data of the first user, a prompt interface can be output to the first user, and the prompt interface is used to prompt the user that the user data is currently being searched. After obtaining the first user's confirmation instruction on the prompt interface, the relevant steps for obtaining the user data of the first user are started, otherwise the process ends.

[0050] 203. Obtain M data assets, and perform feature engineering processing on the M data assets respectively according to the data asset feature set to obtain M data asset feature data.

[0051] The method of obtaining M data assets is similar to the method of obtaining the user data of the first user in step 102. Please refer to the relevant description of obtaining the user data of the first user in step 102, which will not be repeated here. Among them, M can be set by empirical value or historical data, which is not limited here, and M is a positive integer. Among them, the data assets can be the data dictionary in the big data platform and each source system, and the data dictionary can be data table information, indicator information, data standard, metadata information, etc. This application takes the asset data in the form of a data table as an example for explanation.

[0052] The data asset feature set includes at least one data asset feature, and the data asset feature can be one or more of the data table feature, business theme feature, data table statistical feature, etc. Among them, the data table feature can include one or more of the asset completion rate, the development team, the business team, the subject domain, the table Chinese, and the field Chinese; the business theme feature can include one or more of the customer name, product name, and organization name; the data table statistical feature can be one or more of the total number of fields, field Chinese coverage, and the corresponding standard number.

[0053] Optionally, the data asset feature set can be determined based on one or more of the domain features of the data asset, the type of the data asset recommendation system, the identifier of the data asset recommendation system, and other information. For example, different data asset domains correspond to different data asset feature sets, and different data asset recommendation system types correspond to different data asset feature sets. This helps improve the efficiency of feature engineering processing, thereby improving the efficiency of recommendations and improving the reliability of recommendations.

[0054] 204. Use the data asset recommendation model to process the data search keywords, user feature data, and M data asset feature data to obtain data asset recommendation probabilities corresponding to the N data assets.

[0055] Among them, N can be set by empirical value or historical data, and N is a positive integer less than or equal to M. The recommendation probability corresponding to one or more data assets in the M data assets can be determined by the data asset recommendation model. For example, the recommendation probability of each data asset can be determined separately, and the recommendation probability of the top N data assets is output in descending order of recommendation probability, that is, the recommendation probability of the top N data assets with the highest recommendation probability is output. Data assets are recommended based on the recommendation probability of the N data assets.

[0056] Before using the data asset recommendation model for data processing, the data asset recommendation model can be trained. For example, in some embodiments, a possible method for obtaining a data asset recommendation model is: obtain multiple sample data, the sample data includes a keyword sample, a user data sample, at least one data asset sample, and a sample label set; according to the first user sample feature set, perform feature engineering on the user data sample to obtain the first feature data; according to the first data asset sample feature set, perform feature engineering on at least one data asset sample to obtain the second feature data corresponding to the at least one data asset sample; use the keyword sample, the first feature data, the second feature data corresponding to the at least one data asset sample, and the sample label set to train the initial recommendation model to obtain the data asset recommendation model. That is, the data asset recommendation model can be trained using the keyword samples, feature data, sample labels, etc. corresponding to each sample data.

[0057] Among them, the sample tag set may include at least one sample tag, such as one data asset sample corresponding to one sample tag. Optionally, the sample tag can be used to indicate whether to respond to the data asset, such as 1 for response and 0 for no response; and / or, the sample tag can be used to indicate the recommendation probability of the data asset. The sample data can be pre-stored historical data, such as stored in a database, cloud platform or blockchain. Optionally, the keyword sample can be a keyword corresponding to the domain information, a keyword corresponding to the preference information, etc., which is not limited here.

[0058] Optionally, the first user sample feature set and the first data asset sample feature set may be pre-set, such as input and set by a backend administrator, or may be obtained from a database, cloud platform, blockchain, etc. Further optionally, the first user sample feature set and the first data asset sample feature set may also be determined based on the domain characteristics of the data asset, such as the first user sample feature set and / or the first data asset sample feature set corresponding to different data asset domains may be different. For another example, the first user sample feature set and / or the first data asset sample feature set may be determined based on one or more of the information such as the type of the data asset recommendation system, the identifier of the data asset recommendation system, etc., which will not be elaborated here.

[0059] According to the first user sample feature set, feature engineering is performed on the user data sample to obtain the first feature data, which can be referred to as the method of performing feature engineering on the user data of the first user according to the user feature set to obtain the user feature data; according to the first data asset sample feature set, feature engineering is performed on at least one data asset sample to obtain the second feature data corresponding to the at least one data asset sample, which can be referred to as the method of performing feature engineering on the user data of the first user according to the user feature set to obtain the user feature data, which will not be repeated here. The first feature data can be used to represent the feature data of the user sample, and the second feature data can be used to represent the feature data of the data asset. Optionally, the first user sample feature set and the user feature set can be the same or different.

[0060] Among them, the data asset recommendation model can be used to determine the recommendation probability of one or more data assets. For example, the data asset recommendation model can be a wide&deep model, which is divided into a WIDE model (generalized linear model) and a DEEP model (such as a DNN deep network model), which is obtained by training two models at the same time and weighting them as the final prediction result. That is, the multiple groups of user feature data, asset feature data, sample labels, keyword samples, etc. can be used as sample data, and can be input into the wide&deep model after onehot encoding for model training.

[0061] The DEEP model includes an Embedding layer and multiple hidden layers (hidden layers). The Wide & Deep model connects the Wide part of a single input layer with the Deep part consisting of an Embedding layer and multiple hidden layers, and inputs them together into the final output layer. The feature set of the Wide part includes the original input features and the transformed features. The transformation includes a cross-product transformation, which can capture the interaction between binary features, and then nonlinearity can be added to the Deep model, i.e., a generalized linear model, and converted into a low-dimensional vector, i.e., an embedding vector. These low-dimensional embedding vectors are input into the hidden layer of the neural network in the forward pass to obtain the final output result. For example, the prediction formula of the data asset recommendation model can be:

[0062]

[0063] Wherein, P(Y=1|x) can be used to represent the probability of recommending asset data to the user. x is feature data, for example, it may include first feature data and second feature data corresponding to at least one data asset sample, Y can be a binary class label, σ(·) is a sigmoid function, φ(x) can be a cross product transformation of the original feature x; W wideis the parameter of the WIDE model, such as the vector of the WIDE model weights; W deep is the weight from the hidden layer to the output layer in the DEEP model, for example, it is applied to the final activation a lf The weight of ; l can be the number of hidden layers, and a can be the result of hidden layer calculation, such as a lf can be the activation of the lth layer, f can be the activation function (such as ReLUs), and b can be the bias value.

[0064] 205. Determine at least one data asset to be pushed to the first user according to the data asset recommendation probabilities corresponding to the N data assets.

[0065] When determining the recommended data assets, in addition to the recommendation probability of the data assets, one or more of the characteristics such as user behavior data and user portraits may also be combined to make data asset recommendations.

[0066] For example, in a possible implementation, according to the data asset recommendation probabilities corresponding to the N data assets, the method for determining at least one data asset to be pushed to the first user may be: sorting the N data assets from large to small according to the recommendation probabilities, determining the sequence numbers corresponding to the N data assets, and determining the top K data assets as the at least one data asset to be pushed to the first user. K may be set by experience or historical data.

[0067] For example, in one possible implementation, a method for recommending data assets to a first user includes: determining a second user matching the first user based on user feature data; obtaining first behavior data of the second user; obtaining a first data asset set corresponding to a positive feedback sequence of the second user based on the first behavior data, and pushing the first data asset set to the first user, the first behavior data including a positive feedback sequence, which is a sequence in which users respond to the pushed data assets.

[0068] Optionally, user feature data may include multiple types of feature data. The method for determining a second user matching a first user may be to determine a user having at least Q feature data of the same type as a second user matching the first user. Q may be set by an experience value or historical data and is not limited here.

[0069] As another example, in one possible implementation, a method for recommending data assets to a first user includes: constructing a user profile of the first user based on the user data of the first user; determining a third user matching the first user based on the user profile, and obtaining second behavior data of the third user; and determining a second set of data assets to be pushed to the first user from N data assets based on the second behavior data of the third user.

[0070] The method for determining the third user matching the first user according to the user portrait may be: determining the user corresponding to the user portrait whose similarity with the user portrait of the first user is higher than a preset similarity threshold as the third user. For example, the preset similarity may be based on the ratio of the number of identical description data in the user portrait of the first user and the user portrait of the third user to the number of user data types in the user portrait of the first user as the preset similarity, and the preset similarity may be set by an experience value or historical data.

[0071] That is to say, in some embodiments, the user's data browsing, search and other behavioral information can also be obtained, and the user's own attributes can be collected as a data set for the user portrait. Then, when performing data recommendation, asset data can be recommended in combination with the user portrait. Optionally, the data set of the user portrait can include job information, affiliated team / department information, context features, user features, etc. For example, after obtaining the recommendation probability of each asset data recommended to the target user, when outputting the recommended asset data to the target user according to the recommendation probability of each asset data, the asset data obtained by the asset data recommendation model can be further screened in combination with the user portrait, such as filtering out the asset data selected / used by users in the same position from the asset data, and filtering out the asset data selected / used by users in the same position from the top N asset data with the highest recommendation probability among the asset data, as the recommended asset data output to the target user, and so on, which are not listed here one by one.

[0072] For example, based on the second behavior data, the data asset set corresponding to the positive feedback sequence of the third user can be determined, and the data asset set includes H data assets. If there are multiple third users, the data asset sets corresponding to the positive feedback sequences of the multiple third users can be obtained, and the data assets corresponding to the positive feedback sequences of the multiple third users are processed as a union to determine the data asset sets corresponding to the multiple third users; the H data assets corresponding to the positive feedback sequence of the third user are processed with the N data assets to obtain the data asset set pushed to the first user, and the data asset set pushed to the first user is determined as the second data asset set. If the number of data assets included in the second data asset set is less than the preset number, L data assets are extracted from the N data assets so that the sum of the number of data assets included in the second data asset set and L is the preset number. The preset number is the minimum value of the number of data assets in the second data asset set. Among them, L can be set by experience value or historical data.

[0073] After determining at least one data asset to be pushed to the first user, information about the at least one data asset may be output to the first user. In addition, when pushing, the first data asset may be pushed to the first user according to a predetermined period, or the first data asset may be pushed to the first user at a predetermined time, or the first data asset may be pushed to the first user after the first user clicks a push button. The predetermined period is set according to the specific situation, and may also be set by the user; the predetermined time may be the time when the user opens the software with a data asset recommendation method in the terminal device.

[0074] In this example, the data search keyword is obtained by identifying the data search information received from the first user; then, based on the user feature set, feature engineering is performed on the obtained user data of the first user to obtain user feature data, and based on the data asset feature set, feature engineering is performed on the M data assets to obtain M data asset feature data; then, the data asset recommendation model is used to process the data search keyword, user feature data, and M data asset feature data to obtain data asset recommendation probabilities corresponding to the N data assets; finally, based on the data asset recommendation probabilities corresponding to the N data assets, at least one data asset to be pushed to the first user is determined, thereby helping to improve the accuracy of at least one data asset pushed to the user.

[0075] In some embodiments, a possible method for obtaining a user feature set may be: determining a first recommendation order for at least one data asset sample based on a sample label set; randomly permuting a feature in the first user sample feature set to obtain a second user sample feature set; performing feature engineering processing on the user data sample based on the second user sample feature set to obtain third feature data; using a data asset recommendation model to process the keyword sample, the third feature data, and the second feature data to obtain a first recommendation probability for the at least one data asset sample; determining a recommendation order for multiple data asset samples based on the first recommendation probability to obtain a second recommendation order; comparing data asset samples with the same serial number in the first recommendation order and the second recommendation order, and counting a first total number of different data asset samples corresponding to the same serial number; when the first total number is less than L, deleting the feature corresponding to the first total number to obtain a user feature set, where L is a positive integer.

[0076] Among them, according to the sample label set, the method for determining the first recommended order of at least one data asset sample can be to sort at least one data asset sample from large to small based on the sample labels in the sample label set, and obtain at least one data asset sorting result, and use the obtained at least one data asset sorting result as the first recommended order. The random replacement method can be to replace a feature in the first user sample feature set with any data, for example, it can be replaced with empty, which is not limited here. According to the second user sample feature set, the method for performing feature engineering on the user data sample to obtain the third feature data can refer to the method for performing feature engineering on the user data of the first user according to the user feature set in step 102 to obtain the user feature data, which is not repeated here. According to the first recommendation probability, the recommended order of multiple data asset samples is determined to obtain the second recommended order. The method can refer to the above-mentioned method for determining the first recommended order of at least one data asset sample according to the sample label set. L can be set by empirical value or historical data, which is not limited here. Optionally, the feature corresponding to the first total number can be the feature before replacement.

[0077] For example, the first recommendation order is [a1, a2, a3, a4, a5, a6, a7], a is used to indicate a data asset sample in at least one data asset sample, L is 3, the first user sample feature set is [b1, b2, b3, b4], and b is used to indicate a feature in the first user sample feature set. Assume that the second user sample feature set is [b1, b2, b3, 0], where 0 represents null. Based on the second user sample feature set, the third feature data can be obtained as [c1, c2, c3], c is used to indicate the feature data corresponding to a feature type in the third feature data, and then the first recommendation probability corresponding to at least one asset data is obtained based on the keyword sample, the third feature data, and the second feature data. Finally, based on the first recommendation probability, at least one data asset is sorted from large to small to obtain the second recommendation order. If the second recommendation order is [a1, a3, a2, a4, a5, a6, a7]. The data assets referred to by the second and third positions in the first and second recommendation orders are different, so the first total number is 2. Since the second total number is less than L, b4 can be deleted, and the obtained user feature set can be [b1, b2, b3].

[0078] In this example, at least one data asset sample is sorted by the sample label set to obtain a first recommendation order; then a feature in the first user sample feature set is replaced to obtain a second user sample feature set, and based on the second user sample feature set, feature engineering is performed on the user data sample to obtain third feature data; the keyword sample, the first feature data, and the third feature data are processed by the data asset recommendation model to obtain a first recommendation probability; based on the first recommendation probability, a second recommendation order can be obtained, and the first recommendation order and the second recommendation order can be used to count the first total number of different data asset samples corresponding to the same serial number; finally, the first total number is judged, and if the first total number is less than L, the feature replaced in the first user sample feature set is deleted, thereby helping to reduce the time for the terminal device to perform feature engineering on M data assets and improve the efficiency of the terminal device operation.

[0079] In some embodiments, a possible method for obtaining a data asset feature set may be: determining a first recommendation order of at least one data asset sample based on a sample label set; randomly replacing a feature in the first data asset sample feature set to obtain a second data asset sample feature set; performing feature engineering processing on the data asset sample based on the second data asset sample feature set to obtain fourth feature data; processing the keyword sample, the first feature data, and the fourth feature data using a data asset recommendation model to obtain a second recommendation probability; determining a recommendation order for multiple data asset samples based on the second recommendation probability to obtain a third recommendation order; comparing the data asset samples with the same sequence number in the first recommendation order and the third recommendation order, and counting the second total number of different data asset samples corresponding to the same sequence number; when the second total number is less than P, deleting the feature corresponding to the second total number to obtain the data asset feature set. P is a positive integer, and P can be set by empirical values ​​or historical data, which is not limited here.

[0080] The method of performing feature engineering on the data asset sample according to the second data asset sample feature set to obtain the fourth feature data can refer to the method of performing feature engineering on the user data of the first user according to the user feature set in step 102 to obtain the user feature data, which will not be repeated here. The method of obtaining the data asset feature set can refer to the aforementioned method of obtaining the user feature set, which will not be repeated here.

[0081] In this example, the first recommendation order is obtained through the sample label set, and then a feature in the first data asset sample feature set is replaced to obtain the second data asset sample feature set, and the data asset sample is feature engineered based on the second data asset sample feature set to obtain the fourth feature data, and then the keyword sample, the first feature data, and the fourth feature data are processed by the data asset recommendation model to obtain the second recommendation probability, and the third recommendation order can be obtained based on the second recommendation probability. The second total number of different data asset samples corresponding to the same serial number can be counted through the first recommendation order and the third recommendation order, and finally the second total number is judged. If the second total number is less than P, the feature replaced in the first data asset sample feature set is deleted, thereby reducing useless features, which helps to reduce the time for terminal devices to perform feature engineering on M data assets and improve the efficiency of terminal device operation.

[0082] In some embodiments, the model parameters such as offset values ​​included in the data asset recommendation model can also be updated to optimize the model. For example, the positive feedback sequence and negative feedback sequence of the first user for the pushed data assets can be obtained, the positive feedback sequence is the sequence in which the user responds to the pushed data assets, and the negative feedback sequence is the sequence in which the user does not respond to the pushed asset data; according to the positive feedback sequence and negative feedback sequence of the first user, the behavior data (user data) of the first user is updated to obtain the behavior update data; according to the behavior update data, the model parameters included in the data asset recommendation model are adjusted to obtain the updated data asset recommendation model. For example, the recommended offset value included in the model parameters can be adjusted to obtain the updated offset value; according to the updated offset value, the data asset recommendation model is optimized to obtain the data asset recommendation optimization model. The data asset recommendation model is optimized by retraining the data asset recommendation model with the data assets recommended by the user based on the data asset recommendation model, so as to further improve the accuracy of the data asset recommendation based on the data asset recommendation model.

[0083] Among them, when obtaining the positive feedback sequence and negative feedback sequence of the first user for the pushed data asset, it can be to detect the operation instruction of the first user in the data asset recommendation interface. If the operation instruction is a response instruction, the relevant information of the data asset corresponding to the response instruction is obtained, and the relevant information of the data asset corresponding to the response instruction is used as a positive feedback sequence; if the operation instruction is an exit instruction or no operation is performed, the relevant information corresponding to the unresponsive data asset set is obtained, and the relevant information corresponding to the unresponsive data asset set is used as a negative feedback sequence. Among them, the operation instruction includes a response instruction and an exit instruction. The response instruction is an instruction for the user to instruct the first user to respond to the recommended data asset, and the exit instruction is an instruction for instructing the first user to exit the data asset recommendation page. Optionally, the relevant information of the data asset can be the name of the data asset, the response time, etc., and can also be the data asset recommendation probability corresponding to the data asset.

[0084] The method for updating the behavior data of the first user based on the positive feedback sequence and negative feedback sequence of the first user obtained above can be as follows: the obtained positive feedback sequence of the first user and the historical positive feedback sequence are processed as a union to obtain the first positive feedback sequence, and the obtained negative feedback sequence of the first user and the historical negative feedback sequence are processed as a union to obtain the first negative feedback sequence; the first positive feedback sequence and the first negative feedback sequence are processed as an intersection to obtain a data set with the same asset data in different feedback sequences; the data set with the same asset data in different feedback sequences in the first positive feedback sequence is deleted to obtain a second positive feedback sequence; the data set with the same asset data in different feedback sequences in the first negative feedback sequence is deleted to obtain a second negative feedback sequence; the second positive feedback sequence and the second negative feedback sequence are combined to obtain behavior update data, thereby completing the update of the behavior data of the first user. For example, the positive feedback sequence of the first user obtained is [e1,e2,e3], and the historical positive feedback sequence is [e6,e7,e8], where e can be the name of the data asset; the negative feedback sequence of the first user obtained is [e4,e5,e6], and the historical negative feedback sequence is [e3,e9,e10], then the first positive feedback sequence is [e1,e2,e3,e6,e7,e8], the first negative feedback sequence is [e4,e5,e6,e3,e9,e10], the data sets of the same asset data in different feedback sequences are [e3,e6], the second positive feedback sequence is [e1,e2,e7,e8], the second negative feedback sequence is [e4,e5,e9,e10], and the behavior update data can be [[e1,e2,e7,e8],[e4,e5,e9,e10]].

[0085] Optionally, the weight value corresponding to the positive feedback sequence can be multiplied by the number of elements contained in the positive feedback sequence to obtain the offset value corresponding to the positive feedback sequence; the weight value corresponding to the negative feedback sequence can be multiplied by the number of elements contained in the negative feedback sequence to obtain the offset value corresponding to the negative feedback sequence; the offset corresponding to the positive feedback sequence and the offset value corresponding to the negative feedback sequence are added together to obtain the updated offset value, thereby completing the update of the offset value. The weight value corresponding to the positive feedback sequence and the weight value corresponding to the negative feedback sequence can be set by empirical values ​​or historical data, which are not limited here. The updated offset value can then be determined as the offset value of the data asset recommendation model to optimize the data asset recommendation model and obtain a data asset recommendation optimization model.

[0086] The present application can identify data search keywords, and then perform feature engineering processing on the user data of the first user obtained based on the user feature set to obtain user feature data, and perform feature engineering processing on M data assets based on the data asset feature set to obtain M data asset feature data, and then use the data asset recommendation model to process the data search keywords, user feature data, and M data asset feature data to obtain the data asset recommendation probabilities corresponding to the N data assets to determine the data assets to be pushed to the first user, which helps to improve the accuracy of data asset recommendations.

[0087] See also Figure 3 , Figure 3 It is a structural diagram of a data asset recommendation device provided by this application. Figure 3 The data asset recommendation device shown can be used to implement the present application Figure 2 For the convenience of explanation, only the part related to the embodiment of the present application is shown. For specific technical details, please refer to the present application. Figure 2 For example, the data asset recommendation device 300 may include: a receiving module 301, an acquisition module 302, a processing module 303, and a data asset recommendation module 304. Among them:

[0088] The receiving module 301 is used to receive data search information input by a first user, identify the data search information, and obtain a data search keyword;

[0089] The acquisition module 302 is used to acquire user data of a first user, and perform feature engineering processing on the user data of the first user according to the user feature set to obtain user feature data; and to acquire M data assets, and perform feature engineering processing on the M data assets respectively according to the data asset feature set to obtain M data asset feature data;

[0090] The processing module 303 is used to process the data search keyword, the user characteristic data, and the M data asset characteristic data using the data asset recommendation model to obtain the data asset recommendation probabilities corresponding to the N data assets respectively;

[0091] The data asset recommendation module 304 is used to determine at least one data asset to be pushed to the first user according to the data asset recommendation probabilities respectively corresponding to the N data assets.

[0092] In a possible implementation, the data asset recommendation device 300 may further include a training module 305 .

[0093] The acquisition module 302 is further used to acquire multiple sample data, the sample data including keyword samples, user data samples, at least one data asset sample, and a sample tag set; the sample tag set includes at least one sample tag, and one data asset sample corresponds to one sample tag; acquire a first user sample feature set and a first data asset sample feature set;

[0094] The processing module 303 is further used to perform feature engineering processing on the user data sample according to the first user sample feature set to obtain first feature data; perform feature engineering processing on at least one data asset sample according to the first data asset sample feature set to obtain second feature data corresponding to the at least one data asset sample;

[0095] The training module 305 is used to train the recommendation model using the keyword sample, the first feature data, the second feature data corresponding to at least one data asset sample, and the sample label set to obtain a data asset recommendation model.

[0096] In a possible implementation, the processing module 303 is further used to determine a first recommendation order of at least one data asset sample according to the sample label set; randomly permute a feature in the first user sample feature set to obtain a second user sample feature set; perform feature engineering processing on the user data sample according to the second user sample feature set to obtain third feature data; use the data asset recommendation model to process the keyword sample, the third feature data, and the second feature data to obtain a first recommendation probability of at least one data asset sample; determine a recommendation order of multiple data asset samples according to the first recommendation probability to obtain a second recommendation order;

[0097] The processing module 303 is also used to compare the data asset samples with the same serial number in the first recommendation order and the second recommendation order, and count the first total number of different data asset samples corresponding to the same serial number; when the first total number is less than L, delete the features corresponding to the first total number in the first user sample feature set to obtain a user feature set, where L is a positive integer.

[0098] In a possible implementation, the processing module 303 is further used to determine a first recommendation order of at least one data asset sample according to the sample tag set; randomly replace a feature in the first data asset sample feature set to obtain a second data asset sample feature set; perform feature engineering processing on the data asset sample according to the second data asset sample feature set to obtain fourth feature data; use the data asset recommendation model to process the keyword sample, the first feature data, and the fourth feature data to obtain a second recommendation probability of at least one data asset sample; determine a recommendation order of multiple data asset samples according to the second recommendation probability to obtain a third recommendation order;

[0099] The processing module 303 is also used to compare the data asset samples with the same serial number in the first recommendation order and the third recommendation order, and count the second total number of different data asset samples corresponding to the same serial number; when the second total number is less than P, delete the features corresponding to the second total number in the first data asset sample feature set to obtain the data asset feature set.

[0100] In a possible implementation, the acquisition module 302 is further configured to determine a second user matching the first user according to the user characteristic data, and acquire the first behavior data of the second user;

[0101] The data asset recommendation module 304 is also used to obtain the first data asset set corresponding to the positive feedback sequence of the second user based on the first behavior data, and push the first data asset set to the first user; the first behavior data includes the positive feedback sequence, and the positive feedback sequence is the sequence in which the user responds to the pushed data assets.

[0102] In a possible implementation, the acquisition module 302 is further configured to construct a user profile of the first user based on the user data of the first user; determine a third user matching the first user based on the user profile, and acquire second behavior data of the third user;

[0103] The data asset recommendation module 304 determines at least one data asset to be output to the first user according to the data asset recommendation probabilities corresponding to the N data assets, and is specifically used to:

[0104] According to the second behavior data of the third user and the data asset recommendation probabilities respectively corresponding to the N data assets, a second data asset set to be pushed to the first user is determined from the N data assets.

[0105] In a possible implementation, the acquisition module 302 is further used to acquire a positive feedback sequence and a negative feedback sequence of the first user to the pushed data asset; the positive feedback sequence is a sequence in which the user responds to the pushed data asset; the negative feedback sequence is a sequence in which the user does not respond to the pushed asset data;

[0106] The processing module 303 is also used to update the behavior data of the first user according to the positive feedback sequence and the negative feedback sequence of the first user to obtain behavior update data; according to the behavior update data, adjust the model parameters included in the data asset recommendation model to obtain an updated data asset recommendation model.

[0107] It can be understood that the various functional modules of the data asset recommendation device of this embodiment can be specifically implemented according to the method in the above method embodiment. The specific implementation process can refer to the relevant description of the above method embodiment, which will not be repeated here.

[0108] See also Figure 4 , Figure 4 It is a structural diagram of a computer device provided by this application, such as Figure 4 As shown, the computer device includes a processor 401 and a memory 402. Optionally, the computer device may further include a network interface 403. The processor 401, the memory 402 and the network interface 403 may exchange data, the network interface 403 is controlled by the processor 401 to send and receive messages, the memory 402 is used to store a computer program, the computer program includes program instructions, and the processor 401 is used to execute the program instructions stored in the memory. The processor 401 is configured to call the program instructions to execute the above method.

[0109] The processor 401 may be a central processing unit (CPU). In one embodiment, the processor 401 may also be a graphics processing unit (GPU). The processor 401 may also be a combination of a CPU and a GPU.

[0110] In a possible implementation, the memory 402 is used to store program instructions, and the processor 401 may call the program instructions to perform the following steps:

[0111] receiving data search information input by a first user, identifying the data search information, and obtaining a data search keyword;

[0112] Acquire user data of the first user, and perform feature engineering processing on the user data of the first user according to the user feature set to obtain user feature data;

[0113] Obtain M data assets, and perform feature engineering on the M data assets according to the data asset feature set to obtain M data asset feature data, where M is a positive integer;

[0114] The data asset recommendation model is used to process the data search keywords, user feature data, and M data asset feature data to obtain the data asset recommendation probabilities corresponding to the N data assets, where N is a positive integer less than or equal to M;

[0115] At least one data asset to be pushed to the first user is determined according to the data asset recommendation probabilities respectively corresponding to the N data assets.

[0116] Optionally, the processor 401 may also be configured to execute:

[0117] Acquire multiple sample data, the sample data including keyword samples, user data samples, at least one data asset sample, and a sample tag set; the sample tag set includes at least one sample tag, and one data asset sample corresponds to one sample tag;

[0118] Obtaining a first user sample feature set and a first data asset sample feature set;

[0119] According to the first user sample feature set, feature engineering processing is performed on the user data sample to obtain first feature data;

[0120] According to the first data asset sample feature set, feature engineering processing is performed on at least one data asset sample to obtain second feature data corresponding to the at least one data asset sample;

[0121] The initial recommendation model is trained using the keyword sample, the first feature data, the second feature data corresponding to at least one data asset sample, and the sample label set to obtain a data asset recommendation model.

[0122] Optionally, the processor 401 may also be configured to execute:

[0123] Determine a first recommendation order of at least one data asset sample according to the sample label set;

[0124] Randomly replace a feature in the first user sample feature set to obtain a second user sample feature set;

[0125] According to the second user sample feature set, feature engineering processing is performed on the user data sample to obtain third feature data;

[0126] Using the data asset recommendation model to process the keyword sample, the third feature data, and the second feature data, to obtain a first recommendation probability of at least one data asset sample;

[0127] Determining a recommendation order of the plurality of data asset samples according to the first recommendation probability to obtain a second recommendation order;

[0128] Compare the data asset samples with the same sequence number in the first recommendation sequence and the second recommendation sequence, and count a first total number of different data asset samples corresponding to the same sequence number;

[0129] When the first total number is less than L, the features corresponding to the first total number in the first user sample feature set are deleted to obtain a user feature set, where L is a positive integer.

[0130] Optionally, the processor 401 may also be configured to execute: determining a first recommendation order of at least one data asset sample according to the sample tag set;

[0131] Randomly replace a feature in the first data asset sample feature set to obtain a second data asset sample feature set;

[0132] According to the second data asset sample feature set, feature engineering processing is performed on the data asset sample to obtain fourth feature data;

[0133] Using the data asset recommendation model to process the keyword sample, the first feature data, and the fourth feature data, to obtain a second recommendation probability of at least one data asset sample;

[0134] Determining a recommendation order of the plurality of data asset samples according to the second recommendation probability to obtain a third recommendation order;

[0135] Compare the data asset samples with the same sequence number in the first recommendation sequence and the third recommendation sequence, and count the second total number of different data asset samples corresponding to the same sequence number;

[0136] When the second total quantity is less than P, the features corresponding to the second total quantity in the first data asset sample feature set are deleted to obtain a data asset feature set, where P is a positive integer.

[0137] Optionally, after the processor 401 determines at least one data asset to be output to the first user according to the data asset recommendation probabilities respectively corresponding to the N data assets, it may also be configured to execute:

[0138] Determining a second user matching the first user based on the user characteristic data;

[0139] Obtaining first behavior data of a second user;

[0140] According to the first behavior data, a first data asset set corresponding to the positive feedback sequence of the second user is obtained, and the first data asset set is pushed to the first user; the first behavior data includes a positive feedback sequence, which is a sequence in which users respond to the pushed data assets.

[0141] Optionally, the processor 401 may also be configured to execute:

[0142] Building a user profile of the first user based on the user data of the first user;

[0143] Determine, based on the user portrait, a third user matching the first user, and obtain second behavior data of the third user;

[0144] Determining at least one data asset to be output to the first user according to the data asset recommendation probabilities respectively corresponding to the N data assets includes:

[0145] According to the second behavior data of the third user and the data asset recommendation probabilities respectively corresponding to the N data assets, a second data asset set to be pushed to the first user is determined from the N data assets.

[0146] Optionally, after the processor 401 determines at least one data asset to be pushed to the first user according to the data asset recommendation probabilities respectively corresponding to the N data assets, it may also be configured to execute:

[0147] Acquire a positive feedback sequence and a negative feedback sequence of the first user to the pushed data asset; the positive feedback sequence is a sequence in which the user responds to the pushed data asset; the negative feedback sequence is a sequence in which the user does not respond to the pushed asset data;

[0148] updating the behavior data of the first user according to the positive feedback sequence and the negative feedback sequence of the first user to obtain behavior update data;

[0149] According to the behavior update data, the model parameters included in the data asset recommendation model are adjusted to obtain an updated data asset recommendation model.

[0150] In a specific implementation, the modules and processor 401 included in the above-described data asset recommendation device can execute part or all of the steps of the data asset recommendation method described in the above-described method embodiment, which will not be repeated here.

[0151] A computer-readable storage medium is also provided in an embodiment of the present application, and the computer-readable storage medium stores a computer program, and the computer program includes program instructions. When the program instructions are executed by the processor, the processor can execute some or all of the steps performed in the above method embodiment. Optionally, the computer storage medium can be volatile or non-volatile. The computer-readable storage medium can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function, etc.; the data storage area can store data created according to the use of the blockchain node, such as user data, data asset information, etc.

[0152] An embodiment of the present application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program enables a computer to execute part or all of the steps of any data asset recommendation method recorded in the above method embodiments.

[0153] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the above-mentioned program can be stored in a computer storage medium, which can be a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the above-mentioned storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0154] The above disclosure is only part of the embodiments of the present application, which certainly cannot be used to limit the scope of rights of the present application. Ordinary technicians in this field can understand that all or part of the processes of implementing the above embodiments and equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A data asset recommendation method, characterized in that: include: receiving data search information input by a first user, identifying the data search information, and obtaining a data search keyword; Acquire user data of the first user, and perform feature engineering processing on the user data of the first user according to the user feature set to obtain user feature data; Obtain M data assets, and perform feature engineering on the M data assets according to the data asset feature set to obtain M data asset feature data, where M is a positive integer; The data search keyword, the user characteristic data, and the M data asset characteristic data are processed using a data asset recommendation model to obtain data asset recommendation probabilities corresponding to the N data assets, where N is a positive integer less than or equal to M; Determining at least one data asset to be pushed to the first user based on the data asset recommendation probabilities corresponding to the N data assets respectively; the method further includes: Building a user profile of the first user based on the user data of the first user; Determine, according to the user portrait, a third user matching the first user, and obtain second behavior data of the third user; The determining, according to the data asset recommendation probabilities respectively corresponding to the N data assets, at least one data asset to be output to the first user comprises: A second data asset set to be pushed to the first user is determined from the N data assets according to the second behavior data of the third user and the data asset recommendation probabilities respectively corresponding to the N data assets.

2. The method according to claim 1, characterized in that: The method further comprises: Acquire multiple sample data, wherein the sample data includes a keyword sample, a user data sample, at least one data asset sample, and a sample tag set; the sample tag set includes at least one sample tag, and one data asset sample corresponds to one sample tag; Obtaining a first user sample feature set and a first data asset sample feature set; According to the first user sample feature set, performing feature engineering processing on the user data sample to obtain first feature data; According to the first data asset sample feature set, feature engineering processing is performed on the at least one data asset sample to obtain second feature data corresponding to the at least one data asset sample; The initial recommendation model is trained using the keyword sample, the first feature data, the second feature data respectively corresponding to the at least one data asset sample, and the sample label set to obtain a data asset recommendation model.

3. The method according to claim 2, characterized in that: The method further comprises: Determining a first recommendation order of the at least one data asset sample according to the sample tag set; Randomly replace a feature in the first user sample feature set to obtain a second user sample feature set; According to the second user sample feature set, performing feature engineering processing on the user data sample to obtain third feature data; Processing the keyword sample, the third feature data, and the second feature data using a data asset recommendation model to obtain a first recommendation probability of the at least one data asset sample; Determining a recommendation order of the plurality of data asset samples according to the first recommendation probability to obtain a second recommendation order; Compare the data asset samples with the same sequence number in the first recommendation sequence and the second recommendation sequence, and count a first total number of different data asset samples corresponding to the same sequence number; When the first total number is less than L, the features corresponding to the first total number in the first user sample feature set are deleted to obtain the user feature set, where L is a positive integer.

4. The method according to claim 2, characterized in that: The method further comprises: Determining a first recommendation order of the at least one data asset sample according to the sample tag set; Randomly replace a feature in the first data asset sample feature set to obtain a second data asset sample feature set; According to the second data asset sample feature set, performing feature engineering processing on the data asset sample to obtain fourth feature data; Processing the keyword sample, the first feature data, and the fourth feature data using a data asset recommendation model to obtain a second recommendation probability of the at least one data asset sample; Determining a recommendation order of the plurality of data asset samples according to the second recommendation probability to obtain a third recommendation order; Compare the data asset samples with the same sequence number in the first recommendation sequence and the third recommendation sequence, and count a second total number of different data asset samples corresponding to the same sequence number; When the second total number is less than P, the features corresponding to the second total number in the first data asset sample feature set are deleted to obtain the data asset feature set, where P is a positive integer.

5. The method according to any one of claims 1 to 4, characterized in that: After determining at least one data asset to be output to the first user according to the data asset recommendation probabilities respectively corresponding to the N data assets, the method further includes: Determining a second user matching the first user according to the user characteristic data; Acquire the first behavior data of the second user; According to the first behavior data, a first data asset set corresponding to the positive feedback sequence of the second user is obtained, and the first data asset set is pushed to the first user; the first behavior data includes a positive feedback sequence, and the positive feedback sequence is a sequence in which users respond to the pushed data assets.

6. The method according to any one of claims 1 to 4, characterized in that: After determining at least one data asset to be pushed to the first user according to the data asset recommendation probabilities respectively corresponding to the N data assets, the method further includes: Acquire a positive feedback sequence and a negative feedback sequence of the first user to the pushed data asset; the positive feedback sequence is a sequence in which the user responds to the pushed data asset; the negative feedback sequence is a sequence in which the user does not respond to the pushed asset data; updating the behavior data of the first user according to the positive feedback sequence and the negative feedback sequence of the first user to obtain behavior update data; According to the behavior update data, the model parameters included in the data asset recommendation model are adjusted to obtain an updated data asset recommendation model.

7. A data asset recommendation device, characterized in that: include: A receiving module receives data search information input by a first user, identifies the data search information, and obtains a data search keyword; an acquisition module, configured to acquire user data of the first user, and perform feature engineering processing on the user data of the first user according to the user feature set to obtain user feature data; acquire M data assets, and perform feature engineering processing on the M data assets respectively according to the data asset feature set to obtain M data asset feature data, where M is a positive integer; A processing module, used to process the data search keyword, the user characteristic data, and the M data asset characteristic data using a data asset recommendation model to obtain data asset recommendation probabilities corresponding to the N data assets, where N is a positive integer less than or equal to M; a data asset recommendation module, configured to determine at least one data asset to be pushed to the first user according to the data asset recommendation probabilities respectively corresponding to the N data assets; The acquisition module is further used to construct a user portrait of the first user based on the user data of the first user; determine a third user matching the first user based on the user portrait, and acquire second behavior data of the third user; The data asset recommendation module determines at least one data asset to be output to the first user according to the data asset recommendation probabilities respectively corresponding to the N data assets, and is specifically configured to: A second data asset set to be pushed to the first user is determined from the N data assets according to the second behavior data of the third user and the data asset recommendation probabilities respectively corresponding to the N data assets.

8. A computer device, characterized in that: The method comprises a processor and a memory, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 6.

Images