Machine learning based information pushing method and device, storage medium, and terminal

Abstract

A kind of information push method and device based on machine learning, storage medium, terminal, the method includes: obtaining target data set, the target data set includes the literature data of multiple alternative users;According to the literature data of multiple alternative users, the feature data of each alternative user is calculated, the feature data includes the first evaluation index of each literature;The feature data of each alternative user is input into trained network model, to obtain the evaluation result of each alternative user;Information push is carried out based on the evaluation result of each alternative user.In the scheme of the present application, the feature data containing first evaluation index is input into trained network model, and the evaluation result of each alternative user is obtained, and information push is carried out according to the obtained evaluation result, which can improve the accuracy of information push.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to an information pushing method and device based on machine learning, a storage medium and a terminal. BACKGROUND

[0002] In the prior art, the academic level or academic value of a user is usually determined by the data such as papers and patents of the user. With the advent of the big data era and the continuous development of science and technology, the data such as papers, patents and scientific research projects is increasing, and the mutual reference between the data of different users is becoming more and more complex.

[0003] Therefore, there is an urgent need for a method capable of more accurately pushing the relevant academic information of a user based on the reference relationship between each literature. SUMMARY

[0004] The technical problem solved by the present application is how to accurately push information based on the academic information such as literature data of a user.

[0005] To solve the above technical problems, the embodiment of the present application provides a kind of information push method based on machine learning, the method comprises: obtaining target data set, the target data set includes the literature data of multiple alternative users;According to the literature data of multiple alternative users, the feature data of each alternative user is calculated, the feature data includes the first evaluation index of each literature;The feature data of each alternative user is input into the trained network model to obtain the evaluation result of each alternative user, the trained network model is obtained by training the preset network model using the feature data and the evaluation result of multiple user samples as training data;Based on the evaluation result of each alternative user, information is pushed;Wherein, the calculation method of the first evaluation index of literature includes: extracting the citation relationship between each literature from the literature data, and determining the citation network according to the citation relationship between each literature, the citation network includes multiple nodes and the connecting line between nodes, the node represents literature, and the connecting line represents the citation relationship between literatures;Determine the center iteration node;Determine the reference node having a citation relationship with the center iteration node in the citation network, and distribute the weight of the reference node to the weight of the center iteration node according to the number of nodes cited by the reference node;Iteratively determine the reference node having a citation relationship with the reference node determined in the last iteration in the citation network, and distribute the weight of the reference node to the weight of the node cited by the reference node according to the number of nodes cited by the reference node determined in this iteration, distribute the weight change value of the reference node whose weight changes in the last iteration to the weight of the node cited by the reference node, when the iteration number reaches the preset value, or the weight change value of the center iteration node falls within the preset numerical range, the weight of the center iteration node is the first evaluation index of the literature corresponding to the center iteration node.

[0006] Optionally, the feature data further includes a second evaluation index, and the calculation method of the second evaluation index includes: extracting the published literature data and the cited literature data of a user according to the literature data, the user is a user sample or an alternative user, the published literature data includes the number of published literatures, and the cited literature data includes the number of cited literatures;The second evaluation index of the user is calculated and determined at least according to the published literature data and the cited literature data.

[0007] Optionally, the feature data further includes a third evaluation index, the literature data includes a time field, and the calculation method of the third evaluation index includes: for each alternative user, filtering out the literature data belonging to a preset time range according to the time field in the literature data;From the literature data belonging to the preset time range, filter out the literature data containing a preset field;Calculate the proportion of the literature data containing the preset field in the literature data belonging to the preset time range to obtain the third evaluation index.

[0008] Optionally, each user sample has a first category identification, the evaluation result has a second category identification, and the method for obtaining the evaluation result of the plurality of user samples comprises: inputting the feature data of each user sample into an initial model to obtain a score of the user sample, the initial model determining the score of each user sample by analytic hierarchy process; classifying the plurality of user samples according to the first category identification to obtain a plurality of user sample sets; determining the ranking of each user sample in the user sample set to which the user sample belongs according to the score of the user sample; and determining the second category identification of the user sample according to the ranking of the user sample in the user sample set to which the user sample belongs.

[0009] Optionally, the method for obtaining the evaluation result of the plurality of user samples further comprises: determining the external terminal matched with each user sample according to the first category identification; sending the literature data of each user sample to the external terminal matched with the user sample, and receiving the third category identification of each user sample from the external terminal; and for each user sample, comparing the second category identification of the user sample with the third category identification, and if the two are different, taking the third category identification as the second category identification of the user sample.

[0010] Optionally, the second category identification comprises a first-level identification, a second-level identification and a third-level identification, and determining the second category identification of each user sample according to the ranking of the user sample in the user sample set to which the user sample belongs comprises: if the ranking of each user sample belongs to a first preset range in the user sample set to which the user sample belongs, determining the second category identification of the user sample as the first-level identification; if the ranking of each user sample is after the first preset range and falls into a second preset range in the user sample set to which the user sample belongs, determining the second category identification of the user sample as the second-level identification; and if the ranking of each user sample is after the second preset range and falls into a third preset range in the user sample set to which the user sample belongs, determining the second category identification of the user sample as the third-level identification.

[0011] Optionally, before inputting the feature data of each candidate user into the trained network model, the method further comprises: obtaining a test data set, the test data set comprising feature data and a first evaluation result of a plurality of test samples; inputting the feature data of each test sample into the trained network model to obtain a second evaluation result; calculating the proportion of the number of test samples whose first evaluation result is consistent with the second evaluation result to the number of all test samples, and if the proportion of the number of test samples whose first evaluation result is consistent with the second evaluation result to the number of all test samples exceeds a preset threshold, determining that the trained network model is accurate.

[0012] To solve the above technical problems, the embodiment of the present application also provides an information pushing device based on machine learning, the device comprising: an acquisition module for acquiring a target data set, the target data set comprising literature data of a plurality of candidate users; a feature determination module for calculating feature data of each candidate user according to the literature data of the plurality of candidate users, the feature data comprising a first evaluation index of each literature; a result calculation module for inputting the feature data of each candidate user into a trained network model to obtain an evaluation result of each candidate user, the trained network model being obtained by training a preset network model by taking feature data and evaluation results of a plurality of user samples as training data; and an information pushing module for pushing information based on the evaluation result of each candidate user; wherein the feature determination module comprises a first feature submodule, the first feature submodule being configured to calculate the first evaluation index of the literature, and the first feature submodule comprising: a citation network construction unit configured to extract citation relationships between literatures from the literature data and determine a citation network according to the citation relationships between the literatures, the citation network comprising a plurality of nodes and connecting lines between the nodes, the nodes representing the literatures, and the connecting lines representing the literatures having citation relationships; a central iteration node determination unit configured to determine a central iteration node; a primary iteration unit configured to determine citation nodes having citation relationships with the central iteration node in the citation network and assign weights of the citation nodes to a weight of the central iteration node according to a number of nodes cited by the citation nodes; and an iteration calculation unit configured to iteratively determine citation nodes having citation relationships with citation nodes determined in a last iteration in the citation network, assign weights of the citation nodes determined in the current iteration to weights of nodes cited by the citation nodes according to a number of nodes cited by the citation nodes determined in the current iteration, assign a weight change value of a citation node whose weight is changed in the last iteration to a weight of a node cited by the citation node, and when a number of iterations reaches a preset value or the weight change value of the central iteration node falls within a preset numerical range, the weight of the central iteration node is the first evaluation index of a literature corresponding to the central iteration node.

[0013] The embodiment of the present application also provides a storage medium having a computer program stored thereon, and the computer program is run on a processor to perform the steps of the information pushing method based on machine learning.

[0014] The embodiment of the present application also provides a terminal comprising a memory and a processor, and the memory has a computer program stored thereon, the computer program being capable of being run on the processor, and the processor performs the information pushing method based on machine learning when the computer program is run.

[0015] Compared with the prior art, the technical scheme of the embodiment of the present application has the following beneficial effects:

[0016] In the scheme of the embodiment of the present application, the citation network is constructed according to the literature data, and the weight of each node in the citation network is calculated according to the citation relationship of each node, the change of the weight of each node will cause the corresponding change of the weight of the cited node, so that the invalid citations such as self-citation and mutual citation can be eliminated, and the first evaluation index (i.e., the weight of the node) obtained can accurately represent the citation situation of each literature, so that the relative importance of the literature can be more accurately evaluated. Then, the network model is trained by using the feature data including the first evaluation index, and the evaluation result of the candidate user is obtained by using the trained network model. Since the first evaluation index can more accurately reflect the citation situation of the literature, the evaluation result obtained is also more accurate, thereby improving the accuracy of the pushed information.

[0017] Further, in the scheme of the embodiment of the present application, the feature data further includes a second evaluation index. Since the second evaluation index is calculated by using the published literature data and the cited literature data, the second evaluation index can reflect the publication quantity and the citation frequency of the candidate user at the same time, avoids the distortion of the result caused by the selection of a single parameter, and makes the evaluation result obtained according to the feature data more real and complete, thereby improving the accuracy of the pushed information. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a flowchart of a kind of information push method based on machine learning in the embodiment of the present application.

[0019] Figure 2 is Figure 1 the flowchart of the first kind of specific step of step S101 in

[0020] Figure 3 is Figure 1 the flowchart of the second kind of specific step of step S102 in

[0021] Figure 4 is part of the flowchart of a kind of information push method based on machine learning in the embodiment of the present application.

[0022] Figure 5 is the structure diagram of a kind of information push device based on machine learning in the embodiment of the present application. DETAILED DESCRIPTION

[0023] As described in the background, a method is needed to accurately push relevant academic information of user based on the citation relationship between literatures.

[0024] To solve the above technical problems, the embodiment of the present application provides a kind of information push method based on machine learning, in the scheme of the embodiment of the present application, in the scheme of the embodiment of the present application, according to the reference network of literature data construction, and according to the reference relationship of each node in reference network, the weight of each node is calculated, the change of each node weight will cause the weight of the node it references to change accordingly, so that invalid reference such as self-reference and mutual reference can be eliminated, the first evaluation index (i.e. the weight of node) obtained can accurately represent the cited condition of each literature, so that the relative importance of literature can be more accurately evaluated. Then the evaluation result of candidate user is obtained by inputting the feature data including the first evaluation index into the trained network model, since the first evaluation index can more accurately reflect the cited condition of literature, the evaluation result obtained is also more accurate, thereby improving the accuracy of information push.

[0025] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0026] Figure 1 is a flow diagram of a kind of information push method based on machine learning in the embodiment of the present application. The method can be executed by terminal, and the terminal can be various appropriate terminal, for example, mobile phone, computer, Internet of Things device, server, etc., but not limited to this. The information pushed can be academic information of multiple candidate users, and the academic information can be user identifier of multiple candidate users, or literature data of at least part of multiple candidate users, but not limited to this.

[0027] Figure 1 The information push method based on machine learning shown can include the following steps:

[0028] Step S101: obtaining a target data set, the target data set includes literature data of multiple candidate users;

[0029] Step S102: calculating feature data of each candidate user according to the literature data of multiple candidate users, the feature data includes first evaluation index of each literature;

[0030] Step S103: inputting the feature data of each candidate user into a trained network model to obtain the evaluation result of each candidate user, the trained network model is obtained by training a preset network model using the feature data and evaluation result of multiple user samples as training data;

[0031] Step S104: information push based on the evaluation result of each candidate user.

[0032] In the implementation of step S101, the terminal can obtain the literature data of the plurality of candidate users. The literature data of the plurality of candidate users can be pre-stored in the terminal locally, or can be stored in an external database, and the terminal can obtain the literature data of the plurality of candidate users by interacting with the external database.

[0033] It should be noted that the literature in the embodiments of the present application can be a paper, a book, a patent document, a conference report, a scientific research project report, or any form of literature, and the embodiments of the present application do not make any limitation thereto.

[0034] Therefore, the literature data in the embodiments of the present application can include the title, abstract, keywords, co-authors, publication, and citation data of the paper published by each candidate user; can also include the name, publishing agency, publishing time, and citation data of the book published by the candidate user, and can also include the bibliographic information of the patent of each candidate user as the inventor or applicant, such as the patent name, applicant, priority information, co-inventor, and the like, and can also include the specification abstract of the patent; and can also include the project name, project leader, project introduction, and project participants of the scientific research project.

[0035] Further, the terminal can perform data cleaning on the obtained literature data of the plurality of candidate users. The process of data cleaning can include operations such as eliminating error data, deduplication, and standardization processing.

[0036] Specifically, each piece of literature data should include the name of the candidate user to which the piece of literature data belongs, such as the author of the paper, the inventor or applicant of the patent, and the project leader of the scientific research project. In the process of data cleaning, the literature data not containing the name of the candidate user can be eliminated.

[0037] In addition, different candidate users can have the same name, and the literature data of the candidate users with the same name needs to be distinguished to avoid the influence of errors on the subsequent processing process without distinction.

[0038] Specifically, the terminal can determine whether candidate users with the same name are the same person based at least on the topic similarity between pairs of document data. Topic similarity can be calculated based on data such as titles, keywords, and abstracts between documents. In the specific calculation, natural language processing algorithms can be used to determine semantic similarity, such as the Chinese Lexical Analysis System (ICTCLAS) of the Institute of Computing Technology, jieba word segmentation, and term frequency–inverse document frequency (TF-IDF) algorithms, etc. This embodiment of the invention does not limit this.

[0039] If the topic similarity between two sets of literature reaches a preset threshold, such as 60%, it can be determined that the two sets of literature belong to the same field, and the candidate users with the same name can be verified as the same person; if the topic similarity is found to be below the preset threshold, it is considered that the candidate users with the same name are not the same person.

[0040] Furthermore, after data cleaning, the terminal can set a user identifier for each candidate user. The user identifier corresponds one-to-one with the candidate user and can uniquely identify the corresponding candidate user. This yields the target dataset, which includes literature data from multiple candidate users, each candidate user having a unique user identifier.

[0041] In the specific implementation of step S102, the characteristic data of each candidate user is calculated based on the literature data of multiple candidate users. The characteristic data refers to data that can be used to indicate the number of publications, the frequency of citations, etc.

[0042] In this embodiment of the invention, the feature data may include a first evaluation index for each document, which can be used to indicate the citation status of each document.

[0043] refer to Figure 2 , Figure 2 It shows Figure 1 A flowchart illustrating the first specific step of step S102. Figure 2 The steps shown can be used to calculate the first evaluation index for each document.

[0044] Figure 2 The method shown for calculating the first evaluation index for each document may include the following steps:

[0045] Step S201: extracting the citation relationship between each article from the literature data, and determining a citation network according to the citation relationship between each article, the citation network comprising a plurality of nodes and connecting lines between the nodes, the nodes representing articles, and the connecting lines representing the citation relationship between articles;

[0046] Step S202: determining a center iteration node;

[0047] Step S203: determining a citation node having a citation relationship with the center iteration node in the citation network, and assigning the weight of the citation node to the weight of the center iteration node according to the number of nodes cited by the citation node;

[0048] Step S204: iteratively determining a citation node having a citation relationship with the citation node determined in the last iteration in the citation network, and assigning the weight of the citation node to the weight of the node cited by the citation node according to the number of nodes cited by the citation node determined in the iteration, assigning the weight change value of the citation node determined in the last iteration to the weight of the node cited by the citation node, and when the iteration number reaches a preset value, or the weight change value of the center iteration node falls within a preset value range, the weight of the center iteration node is the first evaluation index of the article corresponding to the center iteration node.

[0049] It should be noted that in the embodiment of the present application, each node represents an article, and the weight of the node is the first evaluation index of the article corresponding to the node. Since the citation network describes the citation relationship between the nodes through the connecting lines, the first evaluation index calculated according to the citation relationship can represent the citation of the article represented by the node by the plurality of articles represented by the nodes in the citation network, thereby reflecting the importance of the article.

[0050] The first evaluation index (i.e., the weight) in the embodiment can also be referred to as a citation article (CA) index, or any other implementable name, and the embodiment of the present application does not limit this.

[0051] In the specific implementation of step S201, since the terminal obtains the citation relationship between the articles, the terminal can construct a citation network for a plurality of articles based on the obtained literature data. The citation network comprises a plurality of nodes and connecting lines between the nodes. Each node in the citation network represents an article, and the connecting lines between the nodes represent the citation relationship between two articles.

[0052] Furthermore, the connecting lines have a direction, with the node the connecting line points to representing the cited node, and the node connected to the other end of the connecting line representing the referencing node. In a non-limiting embodiment of the present invention, before constructing the citation network, the field to which each candidate user belongs can be determined based on the literature data of each candidate user, and the literature data of that field can be obtained as the literature data for calculating the first evaluation index. That is, the literature data used to construct the citation network may include literature data that does not belong to the candidate users. Therefore, literature with direct or indirect citation relationships to the candidate users' literature can be used to construct the citation network, thereby making the calculated first evaluation index more realistic and accurate in reflecting the citation status of the candidate users' literature, and thus reflecting the importance of the candidate users' literature.

[0053] In a specific implementation of step S202, a central iteration node is determined from among multiple nodes in the reference network. The central iteration node can be randomly selected, or it can be a designated node, etc., and this embodiment of the invention does not impose any restrictions on this.

[0054] In the specific implementation of steps S203 and S204, the terminal performs multiple iterative calculations, and the weight of the central iteration node is updated once after each iteration.

[0055] In the specific implementation of step S203, the terminal performs the first iterative calculation.

[0056] Specifically, firstly, the referencing nodes in the referencing network that have a referencing relationship with the central iteration node are identified. Then, the weights of the referencing nodes are allocated to the weights of the central iteration node based on the number of nodes referenced by each referencing node. Identifying the referencing nodes that have a referencing relationship with the central iteration node means identifying other nodes in the referencing network that reference the central iteration node.

[0057] In the specific implementation of step S204, the terminal performs subsequent multiple iterative calculations until the number of iterations reaches a preset value, or the weight change value of the central iterative node falls into a preset numerical range.

[0058] Specifically, in each subsequent iteration, the reference nodes that have a reference relationship with the reference nodes determined in the previous iteration are first identified in the reference network. Then, the weight of the reference node is assigned to the weight of the reference node it references based on the number of nodes it references in the current iteration. Finally, the weight change value of the reference node whose weight has changed in the previous iteration is assigned to the weight of the reference node it references.

[0059] Furthermore, in each iteration, if the referencing node refers to only a single node, then the weight of the referencing node is entirely allocated to the weight of the node it refers to. That is, when a referencing node refers to only one node, all the weight of the referencing node is allocated to the node it refers to. For example, if the weight of a referencing node is 1, and it refers to only one node (such as node A), then its weight 1 is entirely allocated to node A.

[0060] In each iteration, if a node references multiple nodes, the ratio of the weight of the referencing node to the number of nodes is calculated, and this ratio is allocated to the weight of each node it references. That is, when a referencing node references multiple nodes, its weight can be averaged across the referenced nodes. For example, if a referencing node has a weight of 1 and references two nodes (such as node A and node B), then the referencing node allocates 0.5 of its weight to node A and 0.5 to node B.

[0061] Furthermore, if the difference between the weight of the central iteration node after the (N+1)th iteration and the weight after the Nth iteration falls within a preset numerical range, the iteration process is stopped; or, if the number of iterations reaches a preset value, the iteration process is forcibly stopped.

[0062] Furthermore, in each iteration, the weight of the referencing node can be determined by the following steps: calculating the difference between the current weight of the referencing node and the weight assigned to the nodes it refers to, as the weight of the node after allocation. Thus, after the referencing node allocates its weight to the nodes it refers to, the weight of the referencing node will also decrease accordingly, ensuring the balance of weight values ​​among nodes in the referencing network.

[0063] In another non-limiting embodiment of the present invention, the weight of the reference node can be determined by the following steps in each iteration: calculate the sum of the weights assigned to the reference nodes whose weights have changed during the current iteration, as the weight change value; and continue to assign the weight change value to the weights of the referenced nodes according to the number of nodes referenced by the reference node whose weight has changed.

[0064] If the node referenced by the node whose weight has changed is not the central iteration node, the weight change value of the referenced node continues to be assigned to the node it references until the referenced node is the central iteration node.

[0065] At this point, the weight of the central iteration node can be determined. The weight change value of the central iteration node is the sum of the weights assigned to the central iteration node after multiple iterations. The weight value of the central iteration node is the sum of its initial weight and the weights assigned to the central iteration node after multiple iterations. That is, the first evaluation index of the literature corresponding to the central iteration node can be determined.

[0066] Furthermore, for each document in the target dataset, in order to calculate its first evaluation index, the node corresponding to the document can be used as the central iteration node in the citation network to which the document belongs, and the above steps can be performed.

[0067] Therefore, the first evaluation index of each document in the target dataset can be obtained. It should be noted that if the document data used to calculate the first evaluation index includes document data that does not belong to the candidate users, then it is not necessary to use the node corresponding to the document that does not belong to the candidate user as the central iteration node.

[0068] In a non-limiting embodiment of the present invention, the feature data may further include a second evaluation index for each candidate user, the second evaluation index being calculated based on the candidate user's posting volume and citation volume.

[0069] refer to Figure 3 , Figure 3 It shows Figure 1 A flowchart illustrating the second specific step of step S102. Figure 3 The steps shown can be used to calculate the second evaluation index for each candidate user.

[0070] Figure 3 The calculation shown to determine the second evaluation index for each candidate user may include the following steps:

[0071] Step S301: Extract the published literature data and cited literature data of the candidate users based on the literature data. The published literature data includes the number of published literatures, and the cited literature data includes the number of cited literatures.

[0072] Step S302: Calculate and determine the second evaluation index of the candidate users based at least on the published literature data and the cited literature data.

[0073] In the specific implementation of step S301, for each candidate user, the published literature data of the candidate user can be determined. The published literature data includes the number of documents published by the candidate user and the data of each document published by the candidate user, such as the title, year, source, institution, keywords and abstract of each document. The cited literature data of each candidate user can also be determined. The cited literature refers to the literature that cites the published literature of the candidate user. The cited literature data may include the number of cited literature, the title, year, source, institution, keywords and abstract of the cited literature.

[0074] In the specific implementation of step S302, the second evaluation index can be calculated using at least the number of published documents and the number of cited documents. The second evaluation index is positively correlated with both the number of published documents and the number of cited documents; the more documents a candidate user publishes, the higher their second evaluation index; the more documents a candidate user cites, the higher their second evaluation index.

[0075] Therefore, the second evaluation result is positively correlated with both the number of publications and the number of citations, thus avoiding the distortion caused by using a single parameter. For candidate users with few publications but high citation counts, the above calculation method can obtain a higher second evaluation index, solving the problem in existing technologies where it is impossible to simultaneously determine the publication data and citation counts of documents based solely on the number of publications or citations. In subsequent steps, the evaluation results calculated based on feature data including the second evaluation index are more realistic and accurate.

[0076] In a non-limiting embodiment of the present invention, step S302 may include the following steps: calculating the total number of cited documents using the cited document data; determining the number of zero-cited documents in the published document data, and calculating the weighted sum of the number of zero-cited documents and the number of remaining documents; and calculating the second evaluation index of the candidate user based on the total number of cited documents and the weighted sum.

[0077] Specifically, some of the candidate users' published works contain zero-citation documents, meaning these documents have no cited references. The impact of zero-citation documents on the candidate users' evaluation results is relatively small. Therefore, a second evaluation index can be calculated by weighting the number of zero-citation documents with the number of remaining documents, thus reducing the influence of zero-citation documents on the second evaluation index. This weighted sum reflects the number of documents published by the candidate users.

[0078] In addition, the total number of cited references can be calculated. This involves summing the number of citations for all publications by each candidate user. Based on this total number of citations and the weighted sum mentioned above, a second evaluation index for that candidate user can be calculated.

[0079] It should be noted that any feasible mathematical operation can be used to calculate the total number of cited documents and the aforementioned weighted sum to obtain the second evaluation index for candidate users. For example, the total number of cited documents and the aforementioned weighted sum can be summed or multiplied directly; or, if the total number of cited documents is large, to avoid its excessive influence on the second evaluation index, operations such as taking its square root, cube root, or logarithm can be performed, and then summed or multiplied with the aforementioned weighted sum, etc. The embodiments of this invention do not impose any restrictions on this.

[0080] In another non-limiting embodiment of the present invention, step S302 may include the following steps: calculating the total number of cited documents using the cited document data; determining the number of zero-cited documents in the published document data, and calculating the weighted sum of the number of zero-cited documents and the remaining documents; calculating the H-index of the candidate user; and combining the H-index with the total number of cited documents and the weighted sum to obtain the second evaluation index of the candidate user. That is, in addition to calculating the total number of cited documents and the above-mentioned weighted sum, the H-index of the candidate user is also calculated, and the second evaluation index of the candidate user is calculated based on the H-index, the total number of cited documents, and the above-mentioned weighted sum. It should be noted that any feasible mathematical operation can be used to operate on the H-index, the total number of cited documents, and the above-mentioned weighted sum to obtain the second evaluation index of the candidate user. The specific implementation method for calculating the H-index can refer to the existing methods, and will not be repeated here in the embodiments of the present invention.

[0081] In a specific implementation of step S103, the terminal can also establish a publication database and a citation database based on the literature data. The publication database includes the literature data published by each candidate user, and the citation database includes the literature data cited by the literature data published by each candidate user. The number of publications and the number of cited literatures for each candidate user are determined based on the publication database and the citation database.

[0082] Specifically, before calculating the second evaluation index, the terminal can split and supplement the acquired literature data to form a publication database and a citation database. The publication database is mainly used to store all literature data published by the candidate users (including title, publication year, source journal, affiliated institution, keywords, abstract, etc.), while the citation database is mainly used to store data on how the candidate users' published literature is cited by other literature, thus enabling the determination of citation data such as the number of cited literature for each candidate user.

[0083] Subsequent steps, such as determining the published literature data of candidate users, can be based on data from the publication database; and when determining the cited literature data of candidate users, they can be based on data from the citation database.

[0084] Continue to refer to Figure 1 In another non-limiting embodiment of the present invention, the feature data may further include a third evaluation index, which can be used to indicate the degree of matching between each candidate user's literature and a preset field. The preset field may be pre-set, and the specific content of the preset field is not limited in this embodiment of the present invention.

[0085] In the specific implementation of step S102, the document data includes a time field, such as the publication year of a paper, the application date of a patent, the publication date of a book, etc., but is not limited to these. The calculation method of the third evaluation index includes: for each candidate user, filtering document data belonging to a preset time range based on the time field in the document data; filtering document data containing the preset field from the academic data belonging to the preset time range; calculating the proportion of academic data containing the preset field to the document data belonging to the preset time range, so as to obtain the third evaluation index.

[0086] Specifically, for each candidate user's literature data, the terminal can filter out literature data belonging to a preset time range. This preset time range can be pre-set, for example, the most recent three years (January 1, 2018 to December 31, 2020). For the filtered literature data belonging to the preset time range, further filtering is performed to select literature data containing preset fields. These preset fields can be fields indicating the main content of the literature, such as "artificial intelligence" or "image processing." The proportion of academic data containing the preset fields to the literature data belonging to the preset time range is calculated to obtain the third evaluation index. The larger the proportion, the more the main content of the candidate user's published literature within the preset time range matches the preset fields.

[0087] In practical applications, the preset field can be determined based on the academic information to be recommended. For example, if the information to be recommended is the academic data of a preset number of candidate users in the field of image processing, the preset field can be set to "image processing." The more relevant the candidate user's published literature is to "image processing," the higher the candidate user's third evaluation index. It should be noted that the feature data in this embodiment of the invention may also include other evaluation indices. For example, it may also include a fourth evaluation index, which can be used to indicate the number of collaborating users of the candidate user and the citation status of the collaborating users' published literature; it may also include a fifth evaluation index, which can be used to indicate the number of valid patents and patent licenses of the candidate user as an inventor or applicant. In the specific implementation of step S103, the feature data of each candidate user is input into the trained network model to obtain the evaluation result of each candidate user. The trained network model is obtained by training a preset network model using the feature data and evaluation results of multiple user samples as training data. The evaluation result may be a score, a category identifier, or a ranking of user samples.

[0088] Specifically, before inputting the feature data of each candidate user into the trained network model, the terminal uses the feature data of multiple user samples and the evaluation results as training data to train the preset network model. Using the feature data and evaluation results of multiple user samples to train the network model allows the trained network model to learn the relationship between the feature data and the evaluation results. Therefore, when the feature data of candidate users is subsequently input into the trained network model, the evaluation results of the candidate users can be obtained. It should be noted that the preset network model can be trained before, before, or after step S102; this embodiment of the invention does not impose any limitations on this.

[0089] Before training the preset network model, the terminal can acquire literature data from multiple user samples and calculate and determine the feature data and evaluation results for each user sample based on the literature data from multiple user samples.

[0090] It should be noted that the specific details of calculating and determining the feature data of each user sample based on the literature data of multiple user samples can be found in the above description of calculating and determining the feature data of candidate users based on the literature data of candidate users, and will not be repeated here.

[0091] refer to Figure 4 , Figure 4 The diagram illustrates a method for determining the evaluation result of each user sample based on literature data from multiple user samples, according to an embodiment of the present invention. The method specifically includes the following steps:

[0092] Step S401: Input the literature data of each user sample into the initial model to obtain the score of the user sample. The initial model calculates and determines the score of each user sample using the analytic hierarchy process.

[0093] Step S402: Classify the multiple user samples according to the first category identifier to obtain multiple user sample sets;

[0094] Step S403: Determine the ranking of each user sample in its own set of user samples based on the score of each user sample, and determine the second category identifier of each user sample based on the ranking of each user sample in its own set of user samples.

[0095] In the specific implementation of step S401, the initial model can be preset, and the score of each user sample can be determined by the analytic hierarchy process.

[0096] In a non-limiting embodiment of the present invention, the initial model may include primary indicators, secondary indicators, and tertiary indicators. The score of each user sample is determined by calculating the scores of multiple primary indicators for that user sample and the preset weights of each primary indicator. The score of each primary indicator is determined by calculating the scores of multiple secondary indicators for that user sample and the preset weights of each secondary indicator. The score of each secondary indicator is determined by calculating the scores of multiple tertiary indicators for that user sample and the preset weights of each tertiary indicator. The score of each tertiary indicator is determined by multiple literature data points for that user sample. Further details on using the analytic hierarchy process (AHP) to calculate the score of each candidate user from the literature data can be found in existing literature and will not be elaborated upon here.

[0097] In the specific implementation of step S402, each user sample has a first category identifier. The first category identifier can be used to indicate the main content of the literature of each user sample, thereby determining the research field to which each user sample belongs. By classifying multiple user samples according to the first category identifier of each user sample, one or more user sample sets can be obtained.

[0098] In the specific implementation of step S403, in each user sample set, the ranking of each user sample within its respective user sample set is determined based on its score; a second category identifier for each user sample is determined based on its ranking within its respective user sample set, thereby indicating the ranking of each user sample within its respective user sample set. It should be noted that the ranking of each user sample within its respective user sample set can also be directly used as the evaluation result for that user sample.

[0099] Therefore, by first dividing multiple user samples into multiple user sample sets based on the second category identifier of the user samples, and then determining the ranking of the user samples in their respective user sample sets, the accuracy of the ranking results of the user samples can be improved, thereby avoiding the influence of other user samples with low relevance to the literature on the ranking results.

[0100] In a non-limiting embodiment of the present invention, the second category identifier may include a first-level identifier, a second-level identifier, and a third-level identifier. If the ranking of each user sample is within a leading first preset range in its user sample set, the second category identifier of the user sample is determined to be a first-level identifier. If the ranking of each user sample is after the first preset range and falls within the second preset range in its user sample set, the second category identifier of the user sample is determined to be a second-level identifier. If the ranking of each user sample is after the second preset range and falls within the third preset range in its user sample set, the second category identifier of the user sample is determined to be a third-level identifier.

[0101] In other words, user samples with a Level 1 identifier rank higher than user samples with Level 2 and Level 3 identifiers in their respective user sample sets, and user samples with a Level 2 identifier rank higher than user samples with a Level 3 identifier. For example, user samples ranking less than 1% in their respective user sample sets can have a Level 1 identifier, user samples ranking between 1% and 15% can have a Level 2 identifier, and user samples ranking between 15% and 85% can have a Level 3 identifier.

[0102] Therefore, user samples belonging to different ranges have different second-category labels. Using multiple different second-category labels as the evaluation results of user samples can make the evaluation results of user samples more intuitive and concise, and can quickly determine the information to be pushed based on the evaluation results, thereby improving the efficiency of information recommendation.

[0103] In a non-limiting embodiment of the present invention, step S403 may further include the following steps: sending the literature data of each user sample to an external terminal matching the user sample, and receiving the third category identifier of each user sample from the external terminal, wherein the external terminal is determined based on the first category identifier of each user sample; for each user sample, comparing the second category identifier of the user sample with the third category identifier, and if the two are different, using the third category identifier as the second category identifier of the user sample.

[0104] The third category identifier, similar to the second category identifier, can be used to indicate the ranking of each user sample in its respective user sample set. The third category identifier is calculated and determined by an external terminal.

[0105] Therefore, the third category identifier determined by the external terminal can be used to verify the second category identifier calculated by the initial model, which improves the accuracy of the evaluation results of user samples and thus improves the accuracy of the trained network model.

[0106] Furthermore, before inputting the feature data of each candidate user into the trained network model, the accuracy of the trained network model can be tested first.

[0107] Specifically, the terminal can acquire a test dataset, which includes feature data of multiple test samples and a first evaluation result; input the feature data of each test sample into the trained network model to obtain a second evaluation result; calculate the proportion of test samples whose first evaluation result matches the second evaluation result to the total number of test samples; if the proportion of test samples whose first evaluation result matches the second evaluation result to the total number of test samples exceeds a preset threshold, then the trained network model is determined to be accurate. The preset threshold can be pre-set.

[0108] If the ratio does not exceed the preset threshold, the terminal can obtain more user samples again and repeat the steps above regarding training the preset network model until the test determines that the trained network model is accurate.

[0109] It should be noted that the specific details of obtaining the feature data of the test samples and the first evaluation results can be found in the relevant content above, and will not be repeated here.

[0110] Continue to refer to Figure 1 If the trained network model is determined to be accurate, the terminal inputs the feature data calculated in step S102 into the trained network model to obtain the evaluation result for each candidate user. Since the trained network model learns the relationship between the feature data and the evaluation result, it can output the evaluation result for each candidate user based on the input feature data.

[0111] The evaluation results of the candidate users are in the same form as the evaluation results of the user samples. If the evaluation results of the user samples are second category identifiers, then the evaluation results of the candidate users are also second category identifiers. If the evaluation results of the user samples are scores, then the output evaluation results of the candidate users are also scores.

[0112] Furthermore, the terminal can store the calculated evaluation results of each candidate user, forming a user database. The user database can include the user identifier of each candidate user, their corresponding literature data, and evaluation results. The data in the user database can also be updated periodically.

[0113] In the specific implementation of step S104, after obtaining the evaluation results of each candidate user, the terminal can push information based on the evaluation results of multiple candidate users.

[0114] Specifically, the information pushed can be the user identifiers of one or more candidate users. For example, the user identifiers of candidate users with primary identifiers can be pushed.

[0115] Furthermore, while pushing the user identifier, the evaluation results of the candidate user can also be pushed at the same time, as well as at least a portion of the candidate user's literature data. However, this is not a limitation. The embodiments of the present invention do not impose any restrictions on the specific content and form of the pushed information.

[0116] refer to Figure 5 This invention also provides an information push device based on machine learning. The device may include: an acquisition module 51, used to acquire a target dataset, the target dataset including academic data of multiple candidate users, the academic data including literature data; a feature determination module 52, used to calculate feature data of each candidate user based on the academic data of multiple candidate users, the feature data including a first evaluation index of each literature; a result calculation module 53, used to input the feature data of each candidate user into a trained network model to obtain the evaluation result of each candidate user, the trained network model being trained by using the feature data and evaluation results of multiple user samples as training data to train a preset network model; and an information push module 54, used to push information based on the evaluation result of each candidate user.

[0117] Specifically, the feature determination module 52 includes a first feature submodule (not shown in the figure), which is used to calculate a first evaluation index of the document. The first feature submodule includes:

[0118] A citation network construction unit (not shown in the figure) is used to extract the citation relationships between documents from the document data and determine the citation network based on the citation relationships between documents. The citation network includes multiple nodes and connecting lines between nodes. The nodes represent documents, and the connecting lines represent citation relationships between documents.

[0119] The central iteration node determination unit (not shown in the figure) is used to determine the central iteration node;

[0120] The initial iteration unit (not shown in the figure) is used to determine the reference nodes that have a reference relationship with the central iteration node in the reference network, and to allocate the weight of the reference node to the weight of the central iteration node according to the number of nodes referenced by the reference node.

[0121] An iterative calculation unit (not shown in the figure) iteratively determines, within the citation network, citation nodes that have citation relationships with citation nodes determined in the previous iteration. Based on the number of nodes referenced by the citation nodes determined in this iteration, it allocates the weight of the citation node to the weight of the nodes it references. It also allocates the weight change values ​​of citation nodes whose weights changed in the previous iteration to the weights of the nodes they reference. When the number of iterations reaches a preset value, or when the weight change value of the central iterative node falls within a preset value range, the weight of the central iterative node becomes the first evaluation index of the literature corresponding to the central iterative node.

[0122] For more information on the working principle, operation method, and beneficial effects of the machine learning-based information push method, please refer to [link / reference needed]. Figures 1 to 4 The relevant descriptions in the document will not be repeated here.

[0123] This invention also discloses a storage medium, which is a computer-readable storage medium storing a computer program thereon, the computer program being executable during runtime. Figures 1 to 4 The steps of the method shown are illustrated. The storage medium may include ROM, RAM, disk, or optical disk, etc. The storage medium may also include non-volatile memory or non-transitory memory, etc.

[0124] This invention also discloses a terminal, which may include a memory and a processor. The memory stores a computer program that can run on the processor. When the processor runs the computer program, it can execute... Figures 1 to 4 The steps of the method shown are as follows. The terminal includes, but is not limited to, mobile phones, computers, tablets, and other terminal devices.

[0125] In this application, the term "terminal" can refer to various forms of terminals, access terminals, user units, user stations, mobile stations, mobile stations (MS), remote stations, remote terminals, mobile devices, user terminals, terminal equipment, wireless communication equipment, user agents, or user devices. Terminal equipment can also be cellular phones, cordless phones, Session Initiation Protocol (SIP) phones, Wireless Local Loop (WLL) stations, Personal Digital Assistants (PDAs), handheld devices with wireless communication capabilities, computing devices, or other processing devices connected to a wireless modem, in-vehicle devices, wearable devices, terminal equipment in future 5G networks, or terminal equipment in future evolved Public Land Mobile Networks (PLMNs), etc. This application does not limit the scope of these terminal devices.

[0126] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article indicates that the preceding and following related objects have an "or" relationship.

[0127] In the embodiments of this application, "multiple" refers to two or more.

[0128] The descriptions of "first," "second," etc., appearing in the embodiments of this application are for illustrative purposes and to distinguish the objects being described. They have no order and do not indicate any special limitation on the number of devices in the embodiments of this application, nor do they constitute any limitation on the embodiments of this application.

[0129] In this application embodiment, "connection" refers to various connection methods such as direct connection or indirect connection to realize communication between devices. This application embodiment does not limit this in any way.

[0130] It should be understood that in the embodiments of this application, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0131] It should also be understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0132] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0133] In the several embodiments provided in this application, it should be understood that the disclosed methods, apparatuses, and systems can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for example, the division of units is merely a logical functional division, and other division methods may exist in actual implementation; for example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0134] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0135] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can be physically comprised separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware or in the form of hardware plus software functional units.

[0136] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute some steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0137] While the present invention has been disclosed above, it is not limited thereto. Any person skilled in the art can make various modifications and alterations without departing from the spirit and scope of the invention; therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. A machine learning-based information push method, characterized in that, The method includes: Obtain the target dataset, which includes literature data from multiple candidate users; The characteristic data of each candidate user is calculated based on the literature data of multiple candidate users, and the characteristic data includes the first evaluation index of each literature. The feature data of each candidate user is input into the trained network model to obtain the evaluation result of each candidate user. The trained network model is obtained by training a preset network model with the feature data and evaluation results of multiple user samples as training data. Information is pushed based on the evaluation results of each candidate user; The calculation method for the first evaluation index of the literature includes: The citation relationships between the documents are extracted from the document data, and a citation network is determined based on the citation relationships between the documents. The citation network includes multiple nodes and connecting lines between the nodes. The nodes represent documents, and the connecting lines represent citation relationships between documents. Determine the central iteration node; In the reference network, identify the reference nodes that have a reference relationship with the central iteration node, and allocate the weight of the reference node to the weight of the central iteration node according to the number of nodes referenced by the reference node. Iteratively, reference nodes that have a reference relationship with the reference nodes determined in the previous iteration are identified in the reference network. The weight of the reference node is assigned to the weight of the reference node it references based on the number of nodes it references in the current iteration. The weight change value of the reference node whose weight changed in the previous iteration is assigned to the weight of the reference node it references. When the number of iterations reaches a preset value, or when the weight change value of the central iteration node falls into a preset value range, the weight of the central iteration node is the first evaluation index of the literature corresponding to the central iteration node.

2. The information push method based on machine learning according to claim 1, characterized in that, The feature data also includes a second evaluation index, the calculation method of which includes: Based on the literature data, the user's published literature data and cited literature data are extracted. The user is a user sample or a candidate user. The published literature data includes the number of published literatures, and the cited literature data includes the number of cited literatures. The user's second evaluation index is calculated based at least on the published literature data and the cited literature data.

3. The information push method based on machine learning according to claim 1, characterized in that, The feature data also includes a third evaluation index, the document data includes a time field, and the calculation method for the third evaluation index includes: For each candidate user, literature data belonging to a preset time range is filtered out based on the time field in the literature data; Filter out literature data containing preset fields from the literature data that falls within the preset time range; The proportion of literature data containing the preset fields to the literature data belonging to the preset time range is calculated to obtain the third evaluation index.

4. The information push method based on machine learning according to claim 1, characterized in that, Each user sample has a first category identifier, and the evaluation result has a second category identifier. The method for obtaining the evaluation results of the multiple user samples includes: The literature data for each user sample is input into the initial model to obtain the score for that user sample. The initial model calculates and determines the score for each user sample using the analytic hierarchy process. The multiple user samples are classified according to the first category identifier to obtain multiple user sample sets; The ranking of each user sample in its own set of user samples is determined based on the score of each user sample, and the second category identifier of each user sample is determined based on the ranking of each user sample in its own set of user samples.

5. The information push method based on machine learning according to claim 4, characterized in that, The method for obtaining the evaluation results of the multiple user samples further includes: The external terminal matching each user sample is determined based on the first category identifier; The literature data for each user sample is sent to an external terminal that matches the user sample, and the third category identifier for each user sample is received from the external terminal. For each user sample, the second category identifier of the user sample is compared with the third category identifier. If they are different, the third category identifier is used as the second category identifier of the user sample.

6. The information push method based on machine learning according to claim 4, characterized in that, The second category identifier includes a primary identifier, a secondary identifier, and a tertiary identifier. The second category identifier for each user sample is determined based on its ranking within its respective user sample set, including: If the ranking of each user sample falls within the leading first preset range in its respective user sample set, then the second category identifier of that user sample is determined as the first-level identifier; If the ranking of each user sample is after the first preset range and falls into the second preset range in its user sample set, then the second category identifier of the user sample is determined as the secondary identifier. If the ranking of each user sample is after the second preset range and falls into the third preset range in its user sample set, then the second category identifier of the user sample is determined to be a third-level identifier.

7. The information push method based on machine learning according to claim 1, characterized in that, Before inputting the feature data of each candidate user into the trained network model, the following steps are also included: Obtain a test dataset, which includes feature data of multiple test samples and a first evaluation result; The feature data of each test sample is input into the trained network model to obtain a second evaluation result; Calculate the proportion of test samples whose first evaluation result is consistent with the second evaluation result to the total number of test samples. If the proportion of test samples whose first evaluation result is consistent with the second evaluation result to the total number of test samples exceeds a preset threshold, then the trained network model is determined to be accurate.

8. An information push device based on machine learning, characterized in that, The device includes: The acquisition module is used to acquire the target dataset, which includes literature data from multiple candidate users. The feature determination module is used to calculate the feature data of each candidate user based on the literature data of multiple candidate users. The feature data includes the first evaluation index of each document. The result calculation module is used to input the feature data of each candidate user into the trained network model to obtain the evaluation result of each candidate user. The trained network model is obtained by training a preset network model with the feature data and evaluation results of multiple user samples as training data. The information push module is used to push information based on the evaluation results of each candidate user; The feature determination module includes a first feature submodule, which is used to calculate a first evaluation index for the document. The first feature submodule includes: A citation network construction unit is used to extract the citation relationships between documents from the document data and determine the citation network based on the citation relationships between documents. The citation network includes multiple nodes and connecting lines between nodes. The nodes represent documents, and the connecting lines represent citation relationships between documents. The central iteration node determination unit is used to determine the central iteration node; The initial iteration unit is used to determine the reference nodes that have a reference relationship with the central iteration node in the reference network, and to allocate the weight of the reference node to the weight of the central iteration node according to the number of nodes referenced by the reference node. The iterative calculation unit iteratively determines, within the citation network, citation nodes that have citation relationships with citation nodes determined in the previous iteration. Based on the number of nodes referenced by the citation nodes determined in this iteration, it allocates the weight of each citation node to the weight of the nodes it references. It also allocates the weight change values ​​of citation nodes whose weights changed in the previous iteration to the weights of the nodes they reference. When the number of iterations reaches a preset value, or when the weight change value of the central iterative node falls within a preset numerical range, the weight of the central iterative node becomes the first evaluation index of the literature corresponding to the central iterative node.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it performs the steps of any one of the machine learning-based information push methods according to claims 1 to 7.

10. A terminal comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor runs the computer program, it executes the machine learning-based information push method according to any one of claims 1 to 7.

Images