Information push model training method and information push method, device and equipment

Abstract

The present application discloses an information push model training method and an information push method, device and equipment. The method includes: obtaining a plurality of preset policy information and generating a plurality of random numbers as a plurality of training samples based on a random function; performing dimensionality reduction processing on the keyword feature values ​​corresponding to the plurality of keywords in each policy information to obtain the policy feature values ​​corresponding to each policy information; inputting the random numbers in the training samples into the information push model, matching the random numbers with the policy feature values ​​corresponding to each policy information using the information push model, and outputting the predicted policy information corresponding to the random numbers; adjusting the model parameters of the information push model according to the policy feature values ​​corresponding to the random numbers and the predicted policy information until the information push model converges to obtain the trained information push model. In this way, the push efficiency of policy information can be improved, the waiting time of users can be reduced, and the user experience can be improved.

Description

Technical Field

[0001] The present application belongs to the field of data processing technology, and in particular relates to an information push model training method and an information push method, device and equipment. Background Art

[0002] With the increase in policies, users need to handle more and more matters. However, most users do not know the handling procedures for various matters. This requires pushing corresponding policy information to users to help them handle various matters.

[0003] In the existing policy system, when users search for the policy information they need, due to the huge amount of policy information data, the efficiency of determining the policy information users need from the massive policy information will be relatively low, and the user will have to wait for a long time, which greatly affects the user experience. Summary of the invention

[0004] The embodiments of the present application provide an information push model training method and an information push method, device and equipment, which can at least solve the problem in the prior art that the efficiency of determining the policy data required by the user from massive policy data is relatively low, the user needs to wait for a long time, and the user experience is affected.

[0005] In a first aspect, an embodiment of the present application provides an information push model training method, the method comprising:

[0006] Acquire multiple preset policy information and generate multiple random numbers as multiple training samples based on a random function, each training sample includes one or more random numbers;

[0007] Performing dimensionality reduction processing on keyword feature values ​​corresponding to multiple keywords in each policy information to obtain policy feature values ​​corresponding to each policy information;

[0008] The random numbers in the training samples are input into the information push model, and the random numbers are matched with the policy feature values ​​corresponding to each policy information using the information push model, and the predicted policy information corresponding to the random numbers is output;

[0009] According to the policy feature values ​​corresponding to the random numbers and the predicted policy information, the model parameters of the information push model are adjusted until the information push model converges to obtain the trained information push model.

[0010] In a second aspect, an embodiment of the present application provides an information push method, the method comprising:

[0011] Receive keywords input by users;

[0012] The first keyword feature value corresponding to the keyword input by the user is input into the information push model, and the target policy information corresponding to the first keyword feature value is pushed to the user using the information push model, wherein the information push model is trained according to the information push model training method shown in any one of the embodiments of the first aspect.

[0013] In a third aspect, an embodiment of the present application provides an information push model training device, the device comprising:

[0014] An acquisition module, used for acquiring a plurality of preset policy information and generating a plurality of random numbers as a plurality of training samples based on a random function, each training sample including one or more random numbers;

[0015] A dimension reduction module is used to perform dimension reduction processing on keyword feature values ​​corresponding to multiple keywords in each policy information to obtain a policy feature value corresponding to each policy information;

[0016] A matching module is used to input the random numbers in the training samples into the information push model, use the information push model to match the random numbers with the policy feature values ​​corresponding to each policy information, and output the predicted policy information corresponding to the random numbers;

[0017] The adjustment module is used to adjust the model parameters of the information push model according to the policy feature values ​​corresponding to the random number and the predicted policy information, until the information push model converges to obtain the trained information push model.

[0018] In a fourth aspect, an embodiment of the present application provides an information push device, the device comprising:

[0019] A receiving module, used for receiving keywords input by a user;

[0020] A push module is used to input a first keyword feature value corresponding to a keyword input by a user into an information push model, and use the information push model to push target policy information corresponding to the first keyword feature value to the user, wherein the information push model is trained according to the information push model training method shown in any one of the embodiments of the first aspect.

[0021] In a fifth aspect, an embodiment of the present application provides an electronic device, the device comprising: a processor and a memory storing computer program instructions;

[0022] When the processor executes the computer program instructions, it implements the information push model training method shown in any one of the embodiments of the first aspect and / or the information push method shown in any one of the embodiments of the second aspect.

[0023] In the sixth aspect, an embodiment of the present application provides a computer storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the information push model training method shown in any one of the embodiments of the first aspect and / or the information push method shown in any one of the embodiments of the second aspect are implemented.

[0024] In the seventh aspect, an embodiment of the present application provides a computer program product. When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the information push model training method shown in any one of the embodiments of the first aspect and / or the information push method shown in any one of the embodiments of the second aspect.

[0025] The information push model training method and information push method, device, equipment, medium and product of the embodiments of the present application can obtain multiple preset policy information and perform dimensionality reduction processing on the keyword feature values ​​corresponding to multiple keywords in each policy information to obtain the policy feature value corresponding to each policy information, and then match the random number in the training sample with the policy feature value corresponding to each policy information to obtain the predicted policy information corresponding to the random number. Since the policy feature value corresponding to the policy information is obtained by performing dimensionality reduction processing on the keyword feature values ​​corresponding to multiple keywords in the policy information, the number of policy feature values ​​can be reduced, so that the time for matching the random number with the policy feature value can be greatly shortened. Correspondingly, the time required for information push using the information push model will also be greatly shortened, thereby improving the push efficiency of policy information, reducing the waiting time of users, and improving user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0027] Figure 1 This is a flow chart of an information push model training method provided by an embodiment of the present application;

[0028] Figure 2 It is a policy characteristic value curve diagram of a policy information provided by an embodiment of the present application;

[0029] Figure 3 A policy characteristic value distribution diagram of policy information provided by an embodiment of the present application;

[0030] Figure 4 This is a flow chart of an information push method provided by an embodiment of the present application;

[0031] Figure 5 This is a schematic diagram of process nodes of a policy fulfillment system provided by an embodiment of the present application;

[0032] Figure 6 This is a structural diagram of a data resource planning provided by an embodiment of the present application;

[0033] Figure 7 It is a structural diagram of an information push model training device provided by an embodiment of the present application;

[0034] Figure 8 It is a structural schematic diagram of an information push device provided by an embodiment of the present application;

[0035] Fig. 9 It is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0036] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present application by illustrating the examples of the present application.

[0037] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0038] In addition, it should be noted that the acquisition, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of national laws and regulations.

[0039] As described in the background technology, with the increase in various policies, users need to handle more and more matters, but most users do not know the handling procedures of various matters, which requires pushing corresponding policy information to users to help users handle various matters.

[0040] In the existing policy system, when users search for the policy information they need, due to the huge amount of policy information data, the efficiency of determining the policy information users need from the massive policy information will be relatively low, and the user will have to wait for a long time, which greatly affects the user experience.

[0041] In addition, in the existing technology, policy information can also be pushed through artificial intelligence. Artificial intelligence is a very broad science, which consists of different fields, such as machine learning, computer vision, etc. One of the main goals of artificial intelligence research is to enable machines to be competent for some complex tasks that usually require human intelligence to complete. However, the application of artificial intelligence in the existing service platform is still only at the level of computer learning, that is, manually setting parameters, so that the calculation program obtains the set keywords (i.e., embedding points) in the database, and then automatically screens these keywords, stores the determined ones in the register, and then pushes the required policy information according to the set characteristics of the natural person or legal person. The application of the current government system completely underestimates the capabilities of artificial intelligence.

[0042] Figure 1 A flow chart of an information push model training method provided by an embodiment of the present application is shown. It should be noted that the information push model training method can be applied to an information push model training device, such as Figure 1 As shown, the information push model training method may include the following steps:

[0043] S110, obtaining a plurality of preset policy information and generating a plurality of random numbers as a plurality of training samples based on a random function;

[0044] S120, performing dimensionality reduction processing on keyword feature values ​​corresponding to multiple keywords in each policy information to obtain a policy feature value corresponding to each policy information;

[0045] S130, inputting the random number in the training sample into the information push model, using the information push model to match the random number with the policy feature value corresponding to each policy information, and outputting the predicted policy information corresponding to the random number;

[0046] S140, adjusting the model parameters of the information push model according to the random number and the policy feature value corresponding to the predicted policy information, until the information push model converges, thereby obtaining a trained information push model.

[0047] In this way, it is possible to obtain multiple preset policy information and perform dimensionality reduction processing on the keyword feature values ​​corresponding to multiple keywords in each policy information to obtain the policy feature value corresponding to each policy information, and then match the random number in the training sample with the policy feature value corresponding to each policy information to obtain the predicted policy information corresponding to the random number. Since the policy feature value corresponding to the policy information is obtained by performing dimensionality reduction processing on the keyword feature values ​​corresponding to multiple keywords in the policy information, the number of policy feature values ​​can be reduced, so that the time for matching random numbers with policy feature values ​​can be greatly shortened. Correspondingly, the time required for information push using the information push model will also be greatly shortened. Therefore, the push efficiency of policy information can be improved, the user's waiting time can be reduced, and the user experience can be improved.

[0048] In S110, training samples may be constructed, each of which may include one or more random numbers. The training samples may be used to train an information push model, which may be used to push policy information corresponding to a keyword feature value of a keyword input by a user to a user. Therefore, the random numbers in the training samples may be input as keyword feature values ​​into the information push model to train the information push model. The random numbers may be generated based on a random function.

[0049] For example, random numbers can be generated in the Numpy library through random functions such as np.random.rand(), np.random.randn(), and np.random.randint(). For np.random.randn(), when there is no parameter in the function brackets, a floating point number is returned. When there is one parameter in the function brackets, an array of rank 1 is returned, which cannot represent vectors and matrices. When there are two or more parameters in the function brackets, an array of corresponding dimensions is returned, which can represent vectors or matrices. For np.random.rand(), the usage method is the same as np.random.randn(). np.random.rand() can return one or a group of random sample values ​​that obey the uniform distribution of "0 to 1". The range of random sample values ​​is (0,1), excluding 1; for np.random.randint(), the parameters in the function brackets can be the minimum value (low), the maximum value (high), the array dimension size and the data type. Generally, the default data type can be np.int, and the return value can be (low, high), including low but excluding high; when high is not filled in, the default range of generated random numbers is (0, low). In addition, you can also use np.random.standard_normal() to generate random numbers. np.random.standard_normal() is similar to np.random.randn(), except that the parameters in the function brackets of np.random.standard_normal() are tuples, while the parameters in the function brackets of np.random.randn() are usually integers. When they are floating-point numbers, they are automatically truncated and converted to integers.

[0050] In addition, multiple preset policy information may be obtained so as to store the feature values ​​of keywords in the policy information into the information push model to be trained.

[0051] In some implementations, training samples can also be constructed by acquiring historical behavior data of users. Specifically, it is possible to widely connect to the systems of government departments such as transportation, health, personnel, public security, civil affairs, housing and construction, finance and taxation, education, and medical care, and to achieve collaborative linkage with existing databases of big data platforms such as population databases, legal person databases, electronic license databases, electronic seal databases, and public credit databases, so as to facilitate real-time retrieval of user data when identifying user access behavior. According to the full amount of user data information such as data sources, data flows, and data applications of the big data platform, the data generated by users in the process of using the government service platform is divided into three categories: user basic data, government business data, and user behavior data, so as to set the direction for the next step of data integration, data analysis, and data application.

[0052] Among them, user basic data can cover all basic information of users. Personal user basic data can include personal basic information (gender, age, ID number, place of origin, mobile phone number, email address, work unit), five insurances and one fund (social security, medical insurance, provident fund, etc.), education experience (graduation school, major, degree), tax credit (tax information, credit report), medical health (physical examination report, outpatient medical record, medical diagnosis), real estate and car property (loan record, repayment record, real estate mortgage), employment and entrepreneurship (personal resume, years of work, professional qualifications), travel (travel records, transportation, catering and accommodation), insurance investment (commercial insurance, financial services), etc. Legal person user basic data includes legal person basic information (legal representative information, industry category, social credit code, establishment time, business scope), patent credit (registered patents, corporate credit), land and real estate (land area, term of use, building area) and qualification license (business license, monopoly license, quality certification), etc.

[0053] Government affairs business data can cover the data and related information generated when users handle business on any government affairs service platform. Personal user government affairs business data can include matter handling (application serial number, declaration time, handling progress, handling status), handling evaluation (personal handling evaluation, service guide evaluation), message comments, interactive communication (service consultation, mailbox), online payment (living expenses, traffic fines), logistics delivery (delivery address, logistics history), collection subscription (subscription matters, subscription time, service guide), material upload, electronic seal, electronic certificate and certification materials, etc. Legal person user basic data includes matter handling (application serial number, declaration time, handling progress, handling status), handling evaluation (legal person handling evaluation, service guide evaluation), message comments, tax payment and refund (tax declaration, tax history), industrial and commercial water and electricity (enterprise map, shareholder information, enterprise annual report), logistics warehousing (delivery address, logistics history, warehousing information), policy consultation, subsidy projects, material upload and patent application, etc.

[0054] User behavior data can cover a variety of active behavior information generated when users log in to any government service platform to handle business, which can specifically include user activity information (startup behavior, login behavior, access channel, access time, source region, source domain name), click behavior information (element click, promotional image (Banner) click, site click volume), browsing behavior information (webpage stay time, bounce rate, return visitor, new visitor, return visit number, number of days between return visits, page browsing, H5 browsing, favorite attention, browsing footprint, drainage number, average browsing time), search behavior information (search terms, related keywords, search times) and user preference information (search habits), etc. In order to facilitate the acquisition of user behavior data, user groups can be created: based on operational motivations, relying on user tags, user behavior information and user business handling information, the user range can be circled and user groups can be created. In addition, user groups can be created manually. User group portrait analysis can also be performed: support data distribution statistics and data display of user groups in the tag dimension; support data statistics and data display of user groups' contribution to operational indicators; support cross-comparison between user groups.

[0055] In some implementations, good data must be able to extract good features to really work. Therefore, feature preprocessing and data cleaning are critical steps, which can often significantly improve the effect and performance of the algorithm, such as normalization, discretization, factorization, missing value processing, and collinearity removal. Filter out significant features and discard non-significant features, which requires machine learning engineers to repeatedly understand the business. This has a decisive impact on many results, which requires the use of relevant techniques for feature validity analysis, such as correlation coefficient, chi-square test, average mutual information, conditional entropy, posterior probability, logistic regression weights, and other methods. Therefore, after constructing the training sample by acquiring the user's historical behavior data, the training sample can also be preprocessed, for example, the data can be reviewed and verified, duplicate information can be deleted, invalid information can be eliminated, and erroneous information can be corrected. The consistency, accuracy, authenticity and availability of the data can be improved through data filtering and correction, and the data quality can be improved, which is convenient for accelerating data processing efficiency.

[0056] In some implementations, in order to obtain policy information more accurately and comprehensively, the above-mentioned acquisition of the preset multiple policy information may specifically include:

[0057] The preset multiple policy information is obtained from the government system through the Application Program Interface (API) gateway.

[0058] Here, policy information can be directly obtained from the government system through the API gateway, and the government system can be the system that publishes policy information.

[0059] In this way, obtaining policy information directly from the government system can make the obtained policy information more accurate and comprehensive.

[0060] Of course, policy information can also be obtained from the Internet through crawlers, which is not limited here.

[0061] Involving S120, the keyword feature values ​​corresponding to multiple keywords in each policy information can be reduced in dimension through a conversion function to obtain the policy feature values ​​corresponding to each policy information. In this way, the number of random variables, that is, policy feature values, can be reduced, and a group of unrelated main variables can be obtained, so that the feature values ​​can play a better role in the machine learning algorithm. Specifically, it can be preset to reduce the dimension of the keyword feature values ​​of several keywords to one policy feature value. For example, it can be preset to reduce the dimension of the keyword feature values ​​of three keywords to one policy feature value. If policy information A includes 6 keywords, two policy feature values ​​corresponding to the policy information A can be obtained after the dimension reduction. If policy information B includes 5 keywords, two policy feature values ​​corresponding to the policy information B can also be obtained after the dimension reduction. In addition, in order to reduce the computational complexity, the feature values ​​can also be converted into smaller values ​​that are more suitable for model calculation.

[0062] For example, Figure 2 As shown in the figure, the policy feature value corresponding to a policy information can be fitted into a straight line. Data determines the upper limit of machine learning results, and the algorithm only approaches this upper limit as much as possible. Data must be representative, otherwise it will inevitably be overfitted. Moreover, for classification problems, data skewness cannot be too serious, and the number of data in different categories should not differ by an order of magnitude. In addition, there must be an assessment of the magnitude of the data, how many samples, how many features, and the degree of memory consumption, to determine whether the memory can be placed during the training process. If not, you have to consider improving the algorithm or using some dimensionality reduction techniques. If the amount of data is too large, then you have to adopt distributed methods.

[0063] Exemplarily, the policy information obtained by connecting various government systems through the API gateway can be compiled into a data set. In the data set, a row of data can be a sample, and a column of data can have a characteristic value. Since some data have target values ​​and some data do not have target values, two data types are constituted: characteristic value + target value, and only characteristic value without target value. The target value can be determined based on the user's historical behavior data. The policy classification of different departments can be used to mark the data with characteristic values, and the above-mentioned statistical information on natural persons or enterprises and the behavioral data of operations on the system can be obtained, and these data can be converted into digital features that can be used for machine learning to process the data.

[0064] In some implementations, clarifying the problem is the first step in machine learning, that is, model training. The training process of machine learning is usually a very time-consuming task. First, it is necessary to clarify what kind of data is obtained and the abstracted problem, whether it is a classification, regression or clustering problem.

[0065] Here, since the policy information required by natural persons and enterprises is usually different, two information push models can be trained based on the policy information corresponding to natural persons and the policy information corresponding to enterprises, respectively, or only one information push model can be trained. Figure 3 As shown, different characteristic value intervals are set for the policy information corresponding to natural persons and the policy information corresponding to enterprises, so as to distinguish the policy information corresponding to natural persons from the policy information corresponding to enterprises.

[0066] In some implementations, in order to facilitate matching of the random number with the policy characteristic value, after the above S120, the method may further include:

[0067] The policy feature values ​​are stored in the information push model in the form of a one-dimensional array and a two-dimensional array respectively.

[0068] Here, the policy characteristic values ​​corresponding to the policy information can be stored in the information push model in the form of a one-dimensional array according to the release time of the policy information; at the same time, the policy characteristic values ​​corresponding to the policy information can also be stored in the information push model in the form of a two-dimensional array. Specifically, the policy characteristic values ​​stored in the form of a one-dimensional array and the policy characteristic values ​​stored in the form of a two-dimensional array can be stored in a container defined by a three-dimensional array, such as: a container of a tabular data structure (DataFrame).

[0069] In this way, storing the policy feature values ​​in the form of a one-dimensional array can facilitate subsequent individual matching, and storing them in the form of a two-dimensional array can facilitate subsequent cross-matching.

[0070] In S130, multiple training samples may be input into the information push model to train the information push model. For each training sample, a random number in the training sample may be input into the information push model, and the random number may be matched with a policy feature value corresponding to each policy information using the information push model, and the predicted policy information corresponding to the random number may be output. Before the training sample is input into the information push model, the parameter weights of the model may be randomly initialized.

[0071] Here, the information push model can be a pandas data structure algorithm. Pandas is a tool based on NumPy, which is created to solve data analysis tasks. Pandas incorporates a large number of libraries and some standard data models, and provides tools that can efficiently operate large data sets and a large number of functions and methods that can quickly and conveniently process data.

[0072] In some implementations, in order to train the information push model to push policy information corresponding to a keyword feature value, when a random number is included in the training sample, the above S130 may specifically include:

[0073] The random numbers in the training samples are input into the information push model, and the information push model is used to match the random numbers with the policy feature values ​​stored in the form of a one-dimensional array, and the predicted policy information corresponding to the random numbers is output.

[0074] Here, for each training sample that includes a random number, the training sample can be input into the information push model, and the information push model can be used to individually match the random number with the policy feature value stored in the form of a one-dimensional array, so as to output the predicted policy information corresponding to a random number in the training sample.

[0075] In this way, by training the information push model with a training sample including a random number, the trained information push model can push policy information corresponding to a keyword feature value.

[0076] In some implementations, in order to enable the training information push model to push policy information corresponding to multiple keyword feature values, when the training sample includes multiple random numbers, the above S130 may specifically include:

[0077] The random numbers in the training samples are input into the information push model, and the information push model is used to match the random numbers with the policy feature values ​​stored in the form of a two-dimensional array, and the predicted policy information corresponding to the random numbers is output.

[0078] Here, for each training sample including multiple random numbers, the training sample can be input into the information push model, and the information push model can be used to cross-match the multiple random numbers with the policy feature values ​​stored in the form of a two-dimensional array, so as to output the predicted policy information corresponding to the multiple random numbers in the training sample.

[0079] In this way, by training the information push model with training samples including a plurality of random numbers, the trained information push model can be enabled to push policy information corresponding to a plurality of keyword feature values.

[0080] In S140, the loss function value of the information push model can be determined according to the policy feature value corresponding to the random number and the predicted policy information. When the loss function value does not meet the preset training stop condition, the model parameters of the information push model are adjusted until the loss function value meets the preset training stop condition, that is, the information push model converges, and the trained information push model is obtained. The training stop condition can be pre-set according to user needs. For example, the training stop condition can be that the loss function of the information push model is less than a certain threshold, or the number of iterations of the information push model training reaches a certain number threshold.

[0081] In some embodiments, after the training of the information push model is completed, the trained information push model can be tested. For example, the learning curve method can be used to test whether the trained information push model is overfitting or underfitting. If there is an overfitting or underfitting problem, it is necessary to train again. The basic tuning idea for overfitting is to increase the amount of data and reduce the complexity of the model. The basic tuning idea for underfitting is to increase the number and quality of eigenvalues ​​and increase the complexity of the model. Then perform error analysis and comprehensively analyze the causes of the errors by observing the error samples. The diagnosed model needs to be tuned, and the tuned new model needs to be re-diagnosed. This is a process of repeated iteration and continuous approximation, which requires continuous attempts to reach the optimal state. Then perform a model fusion trial run.

[0082] Here, a sample data set may be obtained in advance and divided into training samples and test samples, for example, in a ratio of 8:2. The sample data set may be a random number generated by a random function, or may be obtained by collecting user historical behavior data such as item click behavior and recommendation lists.

[0083] In some examples, after obtaining test samples by collecting user historical behavior data such as item click behavior and recommendation lists, the item recommendation effect can be analyzed through multiple dimensions to guide product iterative optimization. The analysis indicators can include the recommended click volume, recommended exposure, recommended click rate, average number of clicks per person, recommended user role analysis and popular item indicators, etc.

[0084] In addition, after the information push model is put into operation, the operation accuracy and error can be adjusted according to the usage situation, and the operation speed (time complexity), resource consumption (space complexity), and stability can also be adjusted to obtain a stable information push model.

[0085] In this way, by testing the information push model, the problem of overfitting or underfitting of the information push model can be avoided.

[0086] Combine the following Figure 4The information push method provided in the embodiment of the present application is described in detail.

[0087] Figure 4 A schematic diagram of a flow chart of an information push method provided by an embodiment of the present application is shown. It should be noted that the information push method can be applied to an information push device, such as Figure 4 As shown, the information push method may include the following steps:

[0088] S410, receiving a keyword input by a user;

[0089] S420: Input a first keyword feature value corresponding to the keyword input by the user into an information push model, and use the information push model to push target policy information corresponding to the first keyword feature value to the user.

[0090] Therefore, by inputting the first keyword feature value corresponding to the keyword input by the user into the information push model, the information push model can be used to push the target policy information corresponding to the first keyword feature value to the user. Since the information push model can be a model trained by the above-mentioned information push model training method, the information push model has a high efficiency in information push. Using the information push model to push policy information to users can improve the push efficiency of policy information, reduce user waiting time, and improve user experience.

[0091] In S410, the user may search for policy information by inputting a keyword, and the information push device may receive the keyword input by the user. The keyword input by the user may be one or more keywords.

[0092] In S420, after receiving the keyword input by the user, the first keyword feature value corresponding to the keyword input by the user can be determined according to the correspondence between the keyword and the feature value, and then the first keyword feature value is input into the information push model, and the target policy information corresponding to the first keyword feature value is pushed to the user by the information push model. The information push model here can be a model trained by the above-mentioned information push model training method. The number of target policy information may be one or more, and of course it may also be 0.

[0093] In some implementations, the user may be a natural person user or an enterprise user. The systems or accounts logged in by natural person users and enterprise users when searching may be different. The information push models used by different accounts or systems may be trained based on different training samples. Therefore, when the user is a natural person user and logs in to the system or account corresponding to the natural person to search, the policy information corresponding to the natural person is pushed. When the user is an enterprise user and logs in to the system or account corresponding to the enterprise to search, the policy information corresponding to the enterprise is pushed. In this way, the required policy information can be pushed to the user in a more targeted manner.

[0094] In some implementations, in order to more accurately push policy information to the user, the above S410 may specifically include:

[0095] The first keyword feature value corresponding to the keyword input by the user is input into the information push model, and the first keyword feature value is matched with the policy feature value corresponding to each policy information using the information push model, and the target policy information corresponding to the first keyword feature value is output.

[0096] Here, when a user inputs a keyword to search for policy information, the first keyword feature value corresponding to the keyword input by the user can be input into the information push model, and the first keyword feature value can be matched with the policy feature value corresponding to each policy information using the information push model, and the target policy information corresponding to the first keyword feature value can be output. Specifically, the process of matching the first keyword feature value with the policy feature value corresponding to each policy information is the same as the process of matching the random number in the training sample with the policy feature value corresponding to each policy information, and for the sake of brevity, it will not be repeated here.

[0097] In this way, through the above process, the target policy information can be determined more accurately, so that the policy information can be pushed to the user more accurately.

[0098] In some implementations, if a user enters multiple keywords for search, there may be a situation where multiple keywords correspond to the same policy information. In order to make the policy information pushed to the user more concise, when the keywords entered by the user include at least the first keyword and the second keyword, the first keyword feature value corresponding to the keyword entered by the user is input into the information push model, and the first keyword feature value is matched with the policy feature value corresponding to each policy information by using the information push model, and the target policy information corresponding to the first keyword feature value is output, which may specifically include:

[0099] Inputting the second keyword feature value corresponding to the first keyword and the third keyword feature value corresponding to the second keyword into the information push model, using the information push model to match the second keyword feature value with the policy feature value corresponding to each policy information, and matching the third keyword feature value with the policy feature value corresponding to each policy information, to obtain the first policy feature value that successfully matches the second keyword feature value, and the second policy feature value that successfully matches the third keyword feature value;

[0100] When the first policy characteristic value and the second policy characteristic value include the same third policy characteristic value, the third policy characteristic value included in the first policy characteristic value or the third policy characteristic value included in the second policy characteristic value is removed to obtain a remaining policy characteristic value.

[0101] Output the target policy information corresponding to the remaining policy feature values.

[0102] Here, the user inputs multiple keywords, including the first keyword and the second keyword, the first keyword corresponds to the second keyword feature value, and the second keyword corresponds to the third keyword feature value. The second keyword feature value and the third keyword feature value are input into the information push model, and the information push model can be used to match the second keyword feature value and the third keyword feature value with the policy feature value corresponding to each policy information, so as to obtain the first policy feature value that successfully matches the second keyword feature value and the second policy feature value that successfully matches the third keyword feature value. The number of the first policy feature value and the second policy feature value may be one or more.

[0103] Since the first policy characteristic value and the second policy characteristic value may have the same policy characteristic value, if the policy information corresponding to the first policy characteristic value and the second policy characteristic value is pushed directly to the user, multiple identical policy information may be pushed. In order to avoid this situation, when the first policy characteristic value and the second policy characteristic value include the same third policy characteristic value, the third policy characteristic value included in the first policy characteristic value or the third policy characteristic value included in the second policy characteristic value may be removed to obtain the remaining policy characteristic value. The remaining policy characteristic value may include the first policy characteristic value and the second policy characteristic value without the third policy characteristic value, or the remaining policy characteristic value may include the second policy characteristic value and the first policy characteristic value without the third policy characteristic value.

[0104] In this way, outputting the target policy information corresponding to the remaining policy characteristic values ​​can make each policy information in the target policy information unique without duplication.

[0105] In this way, the pushed policy information can be made more concise, which is convenient for users to view quickly and avoids users wasting time by viewing repeated policy information.

[0106] In some implementations, in order to make the pushed policy information more accurate, before outputting the target policy information corresponding to the remaining policy characteristic value, the method may further include:

[0107] The target policy information corresponding to the remaining policy characteristic values ​​is obtained from the government system through the API gateway.

[0108] Here, after determining the remaining policy characteristic values, the target policy information corresponding to the remaining policy characteristic values ​​can be directly called from the government affairs system through the API gateway.

[0109] In this way, since the government affairs system is a system for publishing policy information, directly using the government affairs system to call the target policy information can make the pushed policy information more accurate.

[0110] In order to more effectively convey the information of various support policies, improve the efficiency of policy implementation, and comprehensively and effectively evaluate the policy effects, local government service data management departments are conducting standardized management of various policy implementation matters, with the goal of achieving one-stop handling of policy implementation matters, realizing "one-stop acceptance, internal circulation, integrated services, and time-limited completion". At the same time, the application matters are tracked and supervised throughout the approval process between various business competent departments to ensure that they are completed within the time limit. It provides good policy implementation services for investors and enterprises; at the same time, it summarizes, counts, and analyzes policy implementation matters to provide analytical decisions and improvement measures for optimizing the business environment. The service objects include specific natural persons and legal persons who meet the policy implementation conditions, and government departments' maintenance personnel for policies and subsidies.

[0111] At present, the existing policy fulfillment service system does not fully apply artificial intelligence. The platform collects, analyzes and pushes policy data through machine learning methods. The level of intelligence is low and it is obviously limited by manual work. There is no reasonable algorithm, which reduces the intelligence level of artificial intelligence. The construction of the policy fulfillment service system is scattered, the concentration is not high, and it is seriously fragmented. There is no unified algorithm and system for the collection, statistics and processing of policy information. The audience is narrow, and the quality of accurate policy push and accurate service needs to be further improved. The various departments implement strip management, there are many barriers between departments, and information cannot be shared, which greatly reduces the sense of gain of enterprises and citizens in enjoying policy services.

[0112] Based on this, the embodiment of the present application also provides a process node diagram of a policy fulfillment system. The information push method in the above embodiment can be applied to the working scenario of the policy fulfillment system. Figure 5 The workflow of the policy implementation system provided in the embodiment of the present application is described in detail.

[0113] Figure 5 A schematic diagram of process nodes of a policy fulfillment system provided by an embodiment of the present application is shown.

[0114] like Figure 5 As shown, the process nodes of the policy fulfillment system may include: an applicant node 510 , an acceptance node 520 , and an approval node 530 .

[0115] Among them, the applicant can fill in the application information in the applicant node 510 to apply online, and then the applicant node 510 and the acceptance node 520 can exchange data, and the application information is sent to the acceptance node 520. If the acceptance result is not passed, the acceptance result can be fed back to the applicant node 510. If the acceptance result is passed, it can enter the approval node 530, and then the result data generated after the approval is completed can be fed back to the applicant node 510.

[0116] In each process node of the policy fulfillment system, applicants can search for any policy information they are not familiar with. The policy fulfillment system can push corresponding policy information to users based on the information push model, thereby providing convenience for applicants in handling matters.

[0117] The present application embodiment also provides a structural diagram of data resource planning. The structure of the data resource planning in the policy fulfillment system provided in the above embodiment can be as follows: Figure 6 As shown, the detailed description is given below.

[0118] Figure 6 A structural diagram of a data resource planning provided by an embodiment of the present application is shown.

[0119] like Figure 6 As shown, the structure of the data resource planning may include: a technical specification security module 610, a data resource layer 620 and a data management and maintenance module 630. The data resource layer 620 may include: a data service 621, a database 622 and a data source 623.

[0120] Among them, data service 621 may include an information resource service module, an information resource directory module, an information resource integration module, and an information resource exchange module;

[0121] Database 622 may include a business database, a history database, a data warehouse, and a metadata database;

[0122] The data source 623 may include a data collection module, a data reporting module and a data exchange module.

[0123] As a result, it is possible to make an overall plan for the data resources of the policy implementation comprehensive service platform at the three levels of collection, storage, and utilization, and establish corresponding technical specifications, data assurance and data security systems, as well as data management and technical maintenance systems.

[0124] Based on the same inventive concept, the present application embodiment also provides an information push model training device. Figure 7 The information push model training device provided in the embodiment of the present application is described in detail.

[0125] Figure 7 A structural schematic diagram of an information push model training device provided by an embodiment of the present application is shown.

[0126] like Figure 7 As shown, the information push model training device may include:

[0127] An acquisition module 701 is used to acquire a plurality of preset policy information and generate a plurality of random numbers as a plurality of training samples based on a random function, each training sample including one or more random numbers;

[0128] A dimension reduction module 702 is used to perform dimension reduction processing on keyword feature values ​​corresponding to multiple keywords in each policy information to obtain a policy feature value corresponding to each policy information;

[0129] A matching module 703 is used to input the random number in the training sample into the information push model, use the information push model to match the random number with the policy feature value corresponding to each policy information, and output the predicted policy information corresponding to the random number;

[0130] The adjustment module 704 is used to adjust the model parameters of the information push model according to the random number and the policy feature value corresponding to the predicted policy information until the information push model converges to obtain the trained information push model.

[0131] In this way, it is possible to obtain multiple preset policy information and perform dimensionality reduction processing on the keyword feature values ​​corresponding to multiple keywords in each policy information to obtain the policy feature value corresponding to each policy information, and then match the random number in the training sample with the policy feature value corresponding to each policy information to obtain the predicted policy information corresponding to the random number. Since the policy feature value corresponding to the policy information is obtained by performing dimensionality reduction processing on the keyword feature values ​​corresponding to multiple keywords in the policy information, the number of policy feature values ​​can be reduced, so that the time for matching random numbers with policy feature values ​​can be greatly shortened. Correspondingly, the time required for information push using the information push model will also be greatly shortened. Therefore, the push efficiency of policy information can be improved, the user's waiting time can be reduced, and the user experience can be improved.

[0132] In some implementations, in order to obtain policy information more accurately and comprehensively, the acquisition module 701 may specifically include:

[0133] The first acquisition submodule is used to obtain multiple preset policy information from the government affairs system through the application programming interface API gateway.

[0134] In some implementations, in order to facilitate matching of the random number with the policy feature value, the apparatus may further include:

[0135] The storage module is used to perform dimensionality reduction processing on the keyword feature values ​​corresponding to multiple keywords in each policy information, obtain the policy feature values ​​corresponding to each policy information, and then store the policy feature values ​​in the form of one-dimensional array and two-dimensional array in the information push model.

[0136] In some implementations, in order to train the information push model to push policy information corresponding to a keyword feature value, when a random number is included in the training sample, the matching module 703 may specifically include:

[0137] The first matching submodule is used to input the random number in the training sample into the information push model, use the information push model to match the random number with the policy feature value stored in the form of a one-dimensional array, and output the predicted policy information corresponding to the random number.

[0138] In some implementations, in order to enable the training information push model to push policy information corresponding to multiple keyword feature values, when the training sample includes multiple random numbers, the matching module 703 may specifically include:

[0139] The second matching submodule is used to input the random numbers in the training samples into the information push model, use the information push model to match the random numbers with the policy feature values ​​stored in the form of a two-dimensional array, and output the predicted policy information corresponding to the random numbers.

[0140] Based on the same inventive concept, the present application embodiment also provides an information push device. Figure 8 The information push device provided in the embodiment of the present application is described in detail.

[0141] Figure 8 A schematic structural diagram of an information push device provided by an embodiment of the present application is shown.

[0142] like Figure 8 As shown, the information push device may include:

[0143] Receiving module 801, used to receive keywords input by users;

[0144] The push module 802 is used to input the first keyword feature value corresponding to the keyword input by the user into the information push model, and use the information push model to push the target policy information corresponding to the first keyword feature value to the user, wherein the information push model is trained by the above-mentioned information push model training method.

[0145] Therefore, by inputting the first keyword feature value corresponding to the keyword input by the user into the information push model, the information push model can be used to push the target policy information corresponding to the first keyword feature value to the user. Since the information push model can be a model trained by the above-mentioned information push model training method, the information push model has a high efficiency in information push. Using the information push model to push policy information to users can improve the push efficiency of policy information, reduce user waiting time, and improve user experience.

[0146] In some implementations, in order to more accurately push policy information to users, the push module 802 may specifically include:

[0147] The third matching submodule is used to input the first keyword feature value corresponding to the keyword input by the user into the information push model, use the information push model to match the first keyword feature value with the policy feature value corresponding to each policy information, and output the target policy information corresponding to the first keyword feature value.

[0148] In some implementations, if a user enters multiple keywords for search, there may be a situation where multiple keywords correspond to the same policy information. In order to make the policy information pushed to the user more concise, when the keywords entered by the user include at least the first keyword and the second keyword, the third matching submodule may specifically include:

[0149] A matching unit, used to input a second keyword feature value corresponding to the first keyword and a third keyword feature value corresponding to the second keyword into an information push model, and use the information push model to match the second keyword feature value with a policy feature value corresponding to each policy information, and to match the third keyword feature value with a policy feature value corresponding to each policy information, to obtain a first policy feature value that successfully matches the second keyword feature value, and a second policy feature value that successfully matches the third keyword feature value;

[0150] A removal unit is used to remove the third policy characteristic value included in the first policy characteristic value or the third policy characteristic value included in the second policy characteristic value when the first policy characteristic value and the second policy characteristic value include the same third policy characteristic value, so as to obtain a remaining policy characteristic value.

[0151] The output unit is used to output the target policy information corresponding to the remaining policy characteristic values.

[0152] In some implementations, in order to make the pushed policy information more accurate, the device may further include:

[0153] The second acquisition submodule is used to obtain the target policy information corresponding to the remaining policy characteristic values ​​from the government affairs system through the API gateway before outputting the target policy information corresponding to the remaining policy characteristic values.

[0154] Fig. 9 A schematic structural diagram of an electronic device provided by an embodiment of the present application is shown.

[0155] like Fig. 9 As shown, the electronic device 9 can implement the exemplary hardware architecture of the electronic device according to the information push model training method and the information push method in the embodiment of the present application, as well as the information push model training device and the information push device. The electronic device can refer to the electronic device in the embodiment of the present application.

[0156] The electronic device 9 may include a processor 901 and a memory 902 storing computer program instructions.

[0157] Specifically, the processor 901 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0158] The memory 902 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 902 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 902 may include a removable or non-removable (or fixed) medium. Where appropriate, the memory 902 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 902 is a non-volatile solid-state memory. In a specific embodiment, the memory 902 may include a read-only memory (ROM), a random access memory (RAM), a disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical, or other physical / tangible memory storage device. Therefore, typically, the memory 902 includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of the present application.

[0159] The processor 901 implements any one of the information push model training methods and / or information push methods in the above embodiments by reading and executing computer program instructions stored in the memory 902.

[0160] In one example, the electronic device may further include a communication interface 903 and a bus 904. Fig. 9 As shown, the processor 901, the memory 902, and the communication interface 903 are connected via a bus 904 and communicate with each other.

[0161] The communication interface 903 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0162] Bus 904 includes hardware, software or both, and the parts of electronic equipment are coupled to each other.For example, but not limitation, bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industrial standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industrial standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In suitable cases, bus 904 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the application considers any suitable bus or interconnection.

[0163] The electronic device can execute the information push model training method and the information push method in the embodiment of the present application, thereby realizing the combination of Figures 1 to 8 Described is an information push model training method, an information push method, a push model training device, and an information push device.

[0164] In addition, in combination with the information push model training method and the information push method in the above embodiments, the present application embodiment can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any one of the information push model training methods and / or information push methods in the above embodiments is implemented.

[0165] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.

[0166] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present application are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0167] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.

[0168] The above reference is according to the method of the embodiment of the present application, the flow chart of the device (system) and the computer program product and / or the block diagram described various aspects of the present application.It should be understood that each square box in the flow chart and / or the block diagram and the combination of each square box in the flow chart and / or the block diagram can be realized by computer program instructions.These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the realization of the function / action specified in one or more square boxes of the flow chart and / or the block diagram.Such a processor can be but is not limited to a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit.It can also be understood that each square box in the block diagram and / or the flow chart and the combination of the square boxes in the block diagram and / or the flow chart can also be realized by the dedicated hardware that performs the specified function or action, or can be realized by the combination of dedicated hardware and computer instructions.

[0169] The above is only a specific implementation of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the protection scope of this application.

Claims

1. A method for training an information push model, characterized in that: The method comprises: Acquire a plurality of preset policy information and generate a plurality of random numbers as a plurality of training samples based on a random function, each of the training samples including one or more random numbers; Performing dimensionality reduction processing on keyword feature values ​​corresponding to multiple keywords in each policy information to obtain a policy feature value corresponding to each policy information; Inputting the random number in the training sample into the information push model, matching the random number with the policy feature value corresponding to each policy information using the information push model, and outputting the predicted policy information corresponding to the random number; According to the random number and the policy feature value corresponding to the predicted policy information, adjusting the model parameters of the information push model until the information push model converges to obtain a trained information push model; In the case where the training sample includes a random number, the random number is input into the information push model, the random number is matched with a policy feature value corresponding to each policy information by using the information push model, and the predicted policy information corresponding to the random number is output, including: Inputting the random number in the training sample into the information push model, matching the random number with the policy feature value stored in the form of a one-dimensional array using the information push model, and outputting the predicted policy information corresponding to the random number; In the case where the training sample includes a plurality of random numbers, the random numbers are input into the information push model, the random numbers are matched with policy feature values ​​corresponding to each of the policy information using the information push model, and the predicted policy information corresponding to the random numbers is output, including: The random numbers in the training samples are input into the information push model, and the random numbers are matched with policy feature values ​​stored in the form of a two-dimensional array using the information push model, and the predicted policy information corresponding to the random numbers is output.

2. The method according to claim 1, characterized in that The obtaining of the preset multiple policy information includes: Obtain multiple preset policy information from the government system through the application programming interface API gateway.

3. The method according to claim 1, characterized in that After performing dimensionality reduction processing on the keyword feature values ​​corresponding to the multiple keywords in each policy information to obtain the policy feature value corresponding to each policy information, the method further includes: The policy characteristic values ​​are stored in the information push model in the form of a one-dimensional array and a two-dimensional array respectively.

4. An information push method, characterized in that: The method comprises: Receive keywords input by users; The first keyword feature value corresponding to the keyword input by the user is input into the information push model, and the target policy information corresponding to the first keyword feature value is pushed to the user using the information push model, wherein the information push model is trained according to the information push model training method according to any one of claims 1-3.

5. The method according to claim 4, characterized in that The step of inputting the first keyword feature value corresponding to the keyword input by the user into the information push model, and using the information push model to push the target policy information corresponding to the first keyword feature value to the user includes: The first keyword feature value corresponding to the keyword input by the user is input into the information push model, and the first keyword feature value is matched with the policy feature value corresponding to each policy information using the information push model to output the target policy information corresponding to the first keyword feature value.

6. The method according to claim 5, characterized in that In the case where the keyword input by the user includes at least a first keyword and a second keyword, inputting a first keyword feature value corresponding to the keyword input by the user into an information push model, matching the first keyword feature value with a policy feature value corresponding to each policy information by using the information push model, and outputting target policy information corresponding to the first keyword feature value, includes: Inputting a second keyword feature value corresponding to the first keyword and a third keyword feature value corresponding to the second keyword into an information push model, using the information push model to match the second keyword feature value with a policy feature value corresponding to each policy information, and matching the third keyword feature value with a policy feature value corresponding to each policy information, to obtain a first policy feature value that successfully matches the second keyword feature value, and a second policy feature value that successfully matches the third keyword feature value; In the case where the first policy characteristic value and the second policy characteristic value include the same third policy characteristic value, removing the third policy characteristic value included in the first policy characteristic value or the third policy characteristic value included in the second policy characteristic value to obtain a remaining policy characteristic value; Output the target policy information corresponding to the remaining policy characteristic value.

7. The method according to claim 6, characterized in that Before outputting the target policy information corresponding to the remaining policy characteristic value, the method further includes: The target policy information corresponding to the remaining policy characteristic values ​​is obtained from the government affairs system through the API gateway.

8. An information push model training device, characterized in that: The device comprises: An acquisition module, used for acquiring a plurality of preset policy information and generating a plurality of random numbers as a plurality of training samples based on a random function, each of the training samples including one or more random numbers; A dimension reduction module, used for performing dimension reduction processing on keyword feature values ​​corresponding to multiple keywords in each policy information to obtain a policy feature value corresponding to each policy information; A matching module, used to input the random number in the training sample into the information push model, use the information push model to match the random number with the policy feature value corresponding to each of the policy information, and output the predicted policy information corresponding to the random number; An adjustment module, used to adjust the model parameters of the information push model according to the random number and the policy feature value corresponding to the predicted policy information, until the information push model converges to obtain a trained information push model; In the case where the training sample includes a random number, the matching module includes: A first matching submodule is used to input the random number in the training sample into the information push model, use the information push model to match the random number with the policy feature value stored in the form of a one-dimensional array, and output the predicted policy information corresponding to the random number; In the case where the training sample includes a plurality of random numbers, the matching module includes: The second matching submodule is used to input the random number in the training sample into the information push model, use the information push model to match the random number with the policy feature value stored in the form of a two-dimensional array, and output the predicted policy information corresponding to the random number.

9. An information push device, characterized in that: The device comprises: A receiving module, used for receiving keywords input by a user; A push module is used to input the first keyword feature value corresponding to the keyword input by the user into the information push model, and use the information push model to push the target policy information corresponding to the first keyword feature value to the user, wherein the information push model is trained according to the information push model training method according to any one of claims 1-3.

10. An electronic device, characterized in that: The device comprises: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the information push model training method described in any one of claims 1-3 and / or the information push method described in any one of claims 4-7.

11. A computer storage medium, characterized in that: The computer storage medium stores computer program instructions, which, when executed by a processor, implement the information push model training method according to any one of claims 1 to 3 and / or the information push method according to any one of claims 4 to 7.

12. A computer program product, characterized in that When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the information push model training method described in any one of claims 1 to 3 and / or the information push method described in any one of claims 4 to 7.

Images