Information recommendation method and device, equipment, storage medium and program product

Abstract

The invention relates to an information recommendation method and device, equipment, a storage medium and a program product. The method comprises the steps that according to historical behavior data of a target user on an information display page, the interest change condition of the target user for various types of information in the financial field is determined, and according to the hit condition of the target user on each leaf node in a user behavior feature tree and the user attribute information and the historical behavior data of the target user, the target user behavior feature tree is determined. The user behavior feature tree is constructed according to the sample behavior data and the sample attribute information, target recommendation information is determined from all the candidate recommendation information according to the interest change situation and the interest statistical situation of the target user for all the types of information in the financial field, and the user behavior feature tree is constructed according to the user behavior data and the sample attribute information. And recommending the target recommendation information to the target user through the information display page. By adopting the method, the reliability of information recommendation can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to an information recommendation method and device, equipment, storage medium and program product. BACKGROUND

[0002] With the continuous development of the financial field, the number of financial information also increases, in order to ensure that the user can obtain the financial information interested by the user when browsing the information display page, the key words in the historical browsing information can be matched in the financial information library, and the matching result is displayed in the information display page.

[0003] However, the way of keyword matching to locate the financial information to be displayed ignores the change of user demand, thereby reducing the reliability of information recommendation. SUMMARY

[0004] Therefore, it is necessary to provide an information recommendation method, device, equipment, storage medium and program product capable of improving the reliability of information recommendation.

[0005] In a first aspect, the present application provides an information recommendation method, comprising:

[0006] According to the historical behavior data of the target user on the information display page, the interest change of the target user on each type of information in the financial field is determined;

[0007] According to the user attribute information and the historical behavior data of the target user, the hit on each leaf node in the user behavior feature tree, the interest statistics of the target user on each type of information in the financial field is determined; wherein the user behavior feature tree is constructed according to the sample behavior data and the sample attribute information;

[0008] According to the interest change and the interest statistics of the target user on each type of information in the financial field, the target recommendation information is determined from each candidate recommendation information, and the target recommendation information is recommended to the target user through the information display page.

[0009] In one embodiment, the interest change includes short-term interest change and long-term interest change; according to the historical behavior data of the target user on the information display page, the interest change of the target user on each type of information in the financial field is determined, comprising:

[0010] The historical behavior data is processed by using deep interest evolution network DIEN to obtain the short-term interest change and the long-term interest change of the target user on each type of information in the financial field; wherein the short-term interest change includes short-term interest type and first interest change trend; the long-term interest change includes long-term interest type and second interest change trend.

[0011] In one of the embodiments, according to the user attribute information and the historical behavior data of the target user, and the hit on each leaf node in the user behavior feature tree, the interest statistical situation of the target user on each type of information in the financial field is determined, including:

[0012] The user attribute information and the historical behavior data of the target user are extracted to obtain user attribute features and user behavior features;

[0013] The user attribute features and the user behavior features are used to traverse each leaf node in the user behavior feature tree to obtain the target leaf node hit;

[0014] The feature information associated with the target leaf node is processed to obtain the interest statistical situation of the target user on each type of information in the financial field.

[0015] In one of the embodiments, according to the interest change situation and the interest statistical situation of the target user on each type of information in the financial field, the target recommendation information is determined from each candidate recommendation information, including:

[0016] According to the information acquisition frequency of the target user on the information display page, an interest adjustment weight is determined;

[0017] According to the interest adjustment weight, the interest change situation and the interest statistical situation of the target user on each type of information in the financial field are processed to obtain an interest score of each type of information;

[0018] According to the interest score of each type of information, the target recommendation information is determined from each candidate recommendation information.

[0019] In one of the embodiments, according to the interest adjustment weight, the interest change situation and the interest statistical situation of the target user on each type of information in the financial field are processed to obtain an interest score of each type of information, including:

[0020] According to the interest change situation of the target user on each type of information in the financial field, a dynamic interest probability corresponding to each type of information is determined;

[0021] According to the interest statistical situation of the target user on each type of information in the financial field, a static interest probability corresponding to each type of information is determined;

[0022] According to the interest adjustment weight, the dynamic interest probability and the static interest probability corresponding to each type of information are processed to obtain an interest score of each type of information.

[0023] In one of the embodiments, the method further includes:

[0024] For each text to be displayed associated with the information display page, semantic parsing is performed on the text to be displayed to obtain the text key fields corresponding to the text to be displayed.

[0025] Input the key text fields into the preset template associated with the information classification model to obtain information analysis instructions;

[0026] Based on the information classification model and information analysis instructions, candidate recommendation information corresponding to the text to be displayed is generated.

[0027] Secondly, this application also provides an information recommendation device, comprising:

[0028] The first determination model is used to determine the changes in target users' interest in various types of information in the financial field based on the target users' historical behavior data on the information display page;

[0029] The second determination model is used to determine the target user's interest statistics on various types of information in the financial field based on the target user's user attribute information and historical behavior data, and the hit status at each leaf node in the user behavior feature tree; wherein, the user behavior feature tree is constructed based on sample behavior data and sample attribute information.

[0030] The information feedback module is used to determine the target recommendation information from the candidate recommendation information based on the target user's changes in interest and interest statistics on various types of information in the financial field, and recommend the target recommendation information to the target user through the information display page.

[0031] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0032] Based on the target users' historical behavior data on the information display page, determine the changes in the target users' interest in various types of information in the financial field;

[0033] Based on the target user's user attribute information and historical behavior data, the hit status at each leaf node in the user behavior feature tree is used to determine the target user's interest statistics on various types of information in the financial field; the user behavior feature tree is constructed based on sample behavior data and sample attribute information.

[0034] Based on the target user's changing interests and interest statistics regarding various types of information in the financial field, target recommended information is determined from all candidate recommended information and then recommended to the target user through the information display page.

[0035] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the following steps:

[0036] According to the historical behavior data of the target user on the information display page, the interest change of the target user on each type of information in the financial field is determined.

[0037] According to the user attribute information and the historical behavior data of the target user, and the hit on each leaf node in the user behavior feature tree, the interest statistics of the target user on each type of information in the financial field is determined; wherein the user behavior feature tree is constructed according to the sample behavior data and the sample attribute information.

[0038] According to the interest change and the interest statistics of the target user on each type of information in the financial field, the target recommended information is determined from each candidate recommended information, and the target recommended information is recommended to the target user through the information display page.

[0039] In a fifth aspect, the present application also provides a computer program product, comprising a computer program, wherein the computer program, when executed by a processor, implements the following steps:

[0040] According to the historical behavior data of the target user on the information display page, the interest change of the target user on each type of information in the financial field is determined.

[0041] According to the user attribute information and the historical behavior data of the target user, and the hit on each leaf node in the user behavior feature tree, the interest statistics of the target user on each type of information in the financial field is determined; wherein the user behavior feature tree is constructed according to the sample behavior data and the sample attribute information.

[0042] According to the interest change and the interest statistics of the target user on each type of information in the financial field, the target recommended information is determined from each candidate recommended information, and the target recommended information is recommended to the target user through the information display page.

[0043] The information recommendation method, device, equipment, storage medium and program product determine the interest statistical situation of the target user according to the interest change situation and the interest statistical situation of the target user on each type of information in the financial field, and determine the target recommendation information from each candidate recommendation information, and recommend the target recommendation information to the target user through the information display page. Compared with the way of positioning the financial information to be displayed by using keyword matching in the related art, the rationality and accuracy of determining the target recommendation information can be ensured by analyzing the historical behavior data of the target user from the two dimensions of interest change and interest statistics by using the above method, thereby improving the reliability of information recommendation. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other related drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0045] Figure 1 The flowchart of the information recommendation method in one embodiment is shown.

[0046] Figure 2 The flowchart of determining the interest statistical situation in one embodiment is shown.

[0047] Figure 3 The flowchart of determining the target recommendation information in one embodiment is shown.

[0048] Figure 4 The flowchart of determining the candidate recommendation information in one embodiment is shown.

[0049] Figure 5 The flowchart of the information recommendation method in another embodiment is shown.

[0050] Figure 6 The structural block diagram of the information recommendation device in one embodiment is shown.

[0051] Figure 7 The internal structure diagram of the computer equipment in one embodiment is shown. DETAILED DESCRIPTION

[0052] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not used to limit the present application.

[0053] With the continuous development of the financial field, the amount of financial information also increases. In order to ensure that the user can obtain the financial information interested by the user when browsing the information display page, the keywords in the historical browsing information can be matched in the financial information library, and the matching result can be displayed in the information display page.

[0054] However, the keyword matching method for positioning the financial information to be displayed ignores the change of user demand, thereby reducing the reliability of information recommendation.

[0055] Therefore, in an exemplary embodiment, an information recommendation method is provided. The method is applied to a server, as shown in Figure 1 The method specifically includes the following steps:

[0056] S101, determining the interest change of the target user to each type of information in the financial field according to the historical behavior data of the target user on the information display page.

[0057] The target user is any registered user in the information display page associated application. The information display page is used to display financial information to the user. The historical behavior data is the behavior data generated when the user performs information acquisition operation on the information display page in the historical period. The historical behavior data can be collected by means of burying points and log collection, and the collected historical behavior data is assigned with a unique behavior identifier and a user identifier, and a behavior time stamp accurate to milliseconds, so that the historical behavior data sequence can be generated according to the user and time sorting in the subsequent.

[0058] Further, the historical behavior data can include but is not limited to basic interaction behavior data (such as browsing, clicking, collecting, sharing, commenting, jumping out), user attribute data (such as user identifier, registration time, region, financial action preference, etc.), context data (such as device information, behavior time) and the like.

[0059] The each type of information is different types of information in the financial field, for example, which can include but is not limited to industry dynamics, economic trade, social livelihood and the like. The interest change is the change of the user's interest degree to each type of information.

[0060] Optionally, the historical behavior data of the target user on the information display page can be input into the trained first interest analysis model, and the first interest analysis model outputs the interest change of the target user to each type of information according to the historical behavior data and the model parameters. The first interest analysis model can be constructed based on a Deep Interest Evolution Network (DIEN).

[0061] S102, according to the user attribute information and the historical behavior data of the target user, and the hit condition of each leaf node in the user behavior feature tree, the interest statistics of the target user to each type of information in the financial field is determined.

[0062] The so-called user attribute information is related information describing the user's attributes, which can include but is not limited to user account, account registration location, etc. The so-called user behavior feature tree is constructed according to sample behavior data and sample attribute information. The so-called interest statistics is the statistical result of the user's interest degree related parameters for each type of information.

[0063] For example, sample behavior features can be extracted from sample behavior data, and sample attribute features can be extracted from sample attribute information. Then, the sample behavior features and sample attribute features can be randomly combined to construct a user behavior feature tree. The leaf node corresponds to a combination feature.

[0064] Optionally, the user attribute information and the historical behavior data of the target user can be feature extracted to obtain each combination feature. Then, each combination feature is used to traverse each leaf node in the user behavior feature tree, so as to obtain the hit condition of each leaf node.

[0065] Further, by counting the hit condition of each leaf node, the interest statistics of the target user to each type of information can be obtained.

[0066] S103, according to the interest change and the interest statistics of the target user to each type of information in the financial field, the target recommendation information is determined from each candidate recommendation information, and the target recommendation information is recommended to the target user through the information display page.

[0067] The so-called candidate recommendation information is un-displayed financial information. Further, the candidate recommendation information includes type information. The so-called target recommendation information is the financial information recommended to the target user.

[0068] Optionally, the comprehensive interest value of the target user for each type of information can be determined in combination with the interest change situation and the interest statistical situation of the target user for each type of information. Then, the comprehensive interest value of each type of information is used to weight the number of information that can be displayed on the information display page, to obtain the displayable information quantity of each type of information. Further, based on the displayable information quantity of each type of information, a corresponding quantity of recommended information is selected from each candidate recommended information as the target recommended information, and is displayed on the information display page.

[0069] In the above information recommendation method, the interest change situation is determined according to the historical behavior data, and the interest statistical situation of the target user is determined according to the user attribute information and the historical behavior data of the target user and the hit situation on each leaf node in the user behavior feature tree. Then, the target recommended information is determined from each candidate recommended information according to the interest change situation and the interest statistical situation of the target user for each type of information in the financial field, and the target recommended information is recommended to the target user through the information display page. Compared with the way of positioning the financial information to be displayed by keyword matching in the related art, the above method can ensure the rationality and accuracy of determining the target recommended information by analyzing the historical behavior data of the target user from the two dimensions of interest change and interest statistics, thereby improving the reliability of information recommendation.

[0070] On the basis of the above embodiment, in the embodiment of the present application, the interest change situation includes a short-term interest change situation and a long-term interest change situation. Based on this, an optional way of determining the interest change situation is provided. Specifically, the historical behavior data is processed by using a deep interest evolution network DIEN to obtain the short-term interest change situation and the long-term interest change situation of the target user for each type of information in the financial field.

[0071] The short-term interest change situation is the interest change situation of the target user in a short time. Further, the short-term interest change situation includes a short-term interest type and a first interest change trend. The short-term interest type is the type information that the target user is interested in in a short time, and the first interest change trend is the interest change trend of the target user for the short-term interest type in a short time.

[0072] The long-term interest change situation is the interest change situation of the target user in a long time. Further, the long-term interest change situation includes a long-term interest type and a second interest change trend. The long-term interest type is the type information that the target user is interested in in a long time, and the second interest change trend is the interest change trend of the target user for the long-term interest type in a long time.

[0073] In order to better analyze the interest change, the historical behavior data can be preprocessed. Specifically, invalid data can be removed. For example, records of user browsing duration < 1 second (false touch) or > 3600 seconds (page hanging); records of the same user clicking the same information > 5 times within 1 minute (malicious clicks); and the same behavior of the same user to the same information (such as repeated collection within 10 minutes) only keep the first record.

[0074] Further, the user behavior sequence can be constructed according to the historical behavior data. For example, group by user identifier, sort the behavior data in ascending order of timestamp, generate a [behavior 1, behavior 2, …, behavior N] sequence (N≤50, take the last 50 behaviors), and convert each behavior data into a triple (information category identifier; behavior type identifier; time interval). Among them, the information category identifier corresponds to the classification result in the first interest analysis model (such as industry dynamics marked as “1”, economic trade marked as “2”, and social livelihood marked as “3”). The behavior type identifier corresponds to each user interaction type (such as browsing = 1, clicking = 2, collecting = 3, sharing = 4, and weight increasing, which can reflect the intensity of interest). The time interval is the time difference between the current behavior and the previous behavior (unit: hour, used to capture interest decay). For users with less than 50 behaviors in the behavior sequence, “empty behavior” (0, 0, 0) can be used to fill in to ensure uniform input length.

[0075] Optionally, the user behavior sequence can be input into the deep interest evolution network DIEN, and the long short-term memory network (LSTM) in the DIEN, i.e., the interest extraction layer, processes the user behavior sequence to obtain the short-term interest type and the first interest change trend. Further, the interest evolution layer can be introduced to distinguish short-term interest and long-term interest by introducing interest units and attention mechanisms, so as to determine the long-term interest type and the second interest change trend.

[0076] Further, the interest change can be integrated by combining the short-term interest type and the first interest change trend, and the long-term interest type and the second interest change trend. Among them, the more upward the interest change trend of each type of information indicates the stronger the current interest in the type of information.

[0077] In the implementation of the present application, by using the DIEN model, the historical behavior data is processed to obtain the interest change, which can ensure the reliability of the determination of the interest change.

[0078] On the basis of the above embodiment, in the embodiment of the present application, an optional way of determining the interest statistical situation is provided, as shown in Figure 2 The specific steps include:

[0079] S201, feature extraction is performed on user attribute information and historical behavior data of a target user to obtain user attribute features and user behavior features.

[0080] The user attribute features are features that can represent user attributes, and the user behavior features are features that can represent user behavior.

[0081] Optionally, static behavior features in the historical behavior data can be extracted to obtain the user attribute features and the user behavior features. Specifically, basic behavior features (such as the total number of behaviors in the last 7 days, the average browsing time, the collection / click ratio, etc.), category preference features, and behavior time features (statistical behavior concentration period (such as whether active in the financial entity processing period (9:30-11:30), represented by 1 / 0) in the historical behavior data can be extracted.

[0082] Further, feature extraction can be performed on the user attribute information, and then the extracted basic attribute features (including user registration duration, registration location), user risk level, and device type features are spliced to obtain the user attribute features.

[0083] S202, the user attribute features and the user behavior features are used to traverse each leaf node in the user behavior feature tree to obtain a hit target leaf node.

[0084] The target leaf node is a leaf node hit by the user features.

[0085] It can be understood that, in order to ensure that the model can capture the association of cross-type features, the user attribute features and the user behavior features can be spliced. Specifically, sequential splicing and type alignment can be used, and the model input features are obtained based on the order of the user attribute features, the basic behavior features, the category preference features, and the behavior time features. For discrete features (such as region and risk level), 0 / 1 values after One-Hot encoding can be used for identification, and for continuous features (such as time length and number of times), [0, 1] values after normalization can be used for identification, and finally a one-dimensional vector with uniform length is formed.

[0086] Optionally, the model input features can be input into a feature tree model to traverse each leaf node in the user behavior feature tree, so as to determine the target leaf node. The feature tree model is a Gradient Boosting Decision Tree (GBDT) model constructed based on the user behavior feature tree.

[0087] S203, processing the feature information associated with the target leaf node to obtain the interest statistics of the target user on each type of information in the financial field.

[0088] Optionally, the feature information associated with the target leaf node can be input into a Logistic Regression (LR) model, and the LR model is used to perform logistic regression calculation on the feature information associated with the target leaf node, so as to obtain the interest statistics of the target user on each type of information. For each type of information, the higher the interest statistical value of the type of information, the stronger the interest in the type of information.

[0089] In the implementation of the present application, by using the GBDT model and the LR model, the historical behavior data and the user attribute information are processed to obtain the interest statistics, which can ensure the reliability of the determination of the interest statistics.

[0090] On the basis of the above-mentioned embodiments, in the embodiments of the present application, an optional way of determining target recommendation information is provided, as shown in Figure 3 The specific steps include the following steps:

[0091] S301, determining an interest adjustment weight according to the information acquisition frequency of the target user on the information display page.

[0092] The so-called information acquisition frequency is the frequency of the user acquiring financial information on the information display page within the sampling period. The so-called interest adjustment weight is used to adjust the proportion of the interest statistics and the interest change.

[0093] It can be understood that in the case of high information acquisition frequency, the interest change occupies a higher proportion; in the case of low information acquisition frequency, the interest statistics occupies a higher proportion.

[0094] Optionally, in an optional implementation, the number of information acquisition times of the target user on the information display page within the sampling period can be counted to determine the information acquisition frequency. In the case that the information acquisition frequency is greater than a frequency threshold, a first weight distribution strategy is used as the interest adjustment weight. In the first weight distribution strategy, the weight corresponding to the interest change is greater than the weight corresponding to the interest statistics.

[0095] In the case that the information acquisition frequency is less than or equal to the frequency threshold, a second weight distribution strategy is used as the interest adjustment weight. In the second weight distribution strategy, the weight corresponding to the interest change is less than the weight corresponding to the interest statistics.

[0096] For example, if the user's recent behavior is frequent, such as the behavior frequency in the last 3 days is more than twice the average in the last 30 days, the interest change condition corresponds to a weight value of 0.7, and the interest statistical condition corresponds to a weight value of 0.3. If the user's recent behavior is sparse, such as the information obtained by the behavior in the last 3 days is less than 5, the interest change condition corresponds to a weight value of 0.3, and the interest statistical condition corresponds to a weight value of 0.7.

[0097] S302, according to the interest adjustment weight, the interest change condition and the interest statistical condition of the target user to each type of information in the financial field are processed to obtain the interest score of each type of information.

[0098] Among them, the so-called interest score is used to represent the degree of interest of the target user to each type of information. The higher the interest score, the higher the degree of interest of the user to the type of information.

[0099] Optionally, for each type of information, the interest change condition and the interest statistical condition of the target user to the type of information can be processed respectively to obtain an initial change value and an initial statistical value. Then, the initial change value and the initial statistical value can be weighted by the interest adjustment weight to obtain the interest score of the type of information.

[0100] For example, according to the interest change condition of the target user to each type of information, the dynamic interest probability corresponding to each type of information is determined; according to the interest statistical condition of the target user to each type of information, the static interest probability corresponding to each type of information is determined; according to the interest adjustment weight, the dynamic interest probability and the static interest probability corresponding to each type of information are processed to obtain the interest score of each type of information.

[0101] Among them, the so-called dynamic interest probability is the degree of interest of the user to the type of information in the dynamic analysis dimension. The so-called static interest probability is the degree of interest of the user to the type of information in the static analysis dimension.

[0102] Optionally, for each type of information, the current interest condition and the interest change trend in the interest change condition of the target user to the type of information can be analyzed to obtain the dynamic interest probability corresponding to the type of information. And, the interest statistical condition of the target user to the type of information can be analyzed to obtain the static interest probability corresponding to the type of information.

[0103] Further, the dynamic interest probability and the static interest probability corresponding to the type of information can be weighted by the interest adjustment weight to obtain the interest score of the type of information.

[0104] S303, according to the interest score of each type of information, the target recommendation information is determined from each candidate recommendation information.

[0105] Optionally, the interest scores of each type of information can be normalized; then, according to the normalized interest scores and the number of information that can be displayed on the information display page, the number of displayable information under each type of information is calculated. Further, for each type of information, the number of displayable information under the type of information can be selected from the candidate recommended information associated with the type of information as the target recommended information.

[0106] It is worth noting that when selecting the candidate recommended information, the candidate recommended information with high access heat can be selected preferentially.

[0107] In the embodiments of the present application, the interest adjustment weight is determined through the information acquisition frequency to process the interest change and the interest statistics, obtain the interest scores of each type of information, and then determine the target recommended information in combination with the interest scores of each type of information, which can ensure the rationality of the target recommended information.

[0108] On the basis of the above-mentioned embodiments, in the embodiments of the present application, an optional way of determining the candidate recommended information is provided, as shown in Figure 4 The method comprises the following steps:

[0109] S401, for each to-be-displayed text associated with the information display page, the to-be-displayed text is subjected to semantic analysis processing to obtain the text key field corresponding to the to-be-displayed text.

[0110] The to-be-displayed text is the to-be-displayed information in the database associated with the to-be-displayed page. For example, it can be financial consulting in the industry database associated with the financial application. The text key field is the field data that can represent the text meaning of the to-be-displayed text.

[0111] Optionally, in order to ensure the real-time performance of the to-be-displayed text acquisition, the to-be-displayed text can be acquired from the preset data source by combining the timing acquisition and the real-time configuration. After the acquisition is completed, the acquired to-be-displayed text can be cleaned and de-duplicated, and the invalid to-be-displayed text can be removed according to the to-be-used scene corresponding to the information display page, so as to obtain the preprocessed to-be-displayed text.

[0112] For each to-be-displayed text, the to-be-displayed text can be subjected to information extraction according to a preset data extraction dimension to obtain the available text corresponding to the to-be-displayed text. The data extraction dimension can include but is not limited to the text title, the core paragraph of the text, and the key entity content (for example, the key financial field).

[0113] Further, the available text can be subjected to semantic analysis, and the text key field corresponding to the to-be-displayed text can be obtained according to the analysis result. For example, the nouns in the financial field can be subjected to standardization processing to avoid ambiguity.

[0114] In addition, for the super-long text, a TextRank algorithm can also be used to compress the super-long text in the process of extracting key sentences in the super-long text according to a preset data extraction dimension; then, the compressed text is subjected to semantic analysis, and the analysis result is subjected to standardization processing to obtain a text key field corresponding to the super-long text. For example, a research report exceeding 2000 words can be compressed into a 500-800 word text key field.

[0115] S402, input the text key field into a preset template associated with the information classification model to obtain an information analysis instruction.

[0116] The preset template is a query instruction template adapted to the information classification model. The information analysis instruction is used to instruct the information classification model to analyze the type information of the to-be-displayed text. The information classification model can be trained by using a sample display text and actual type information corresponding to the sample display text.

[0117] Optionally, to ensure the reliability of the information classification model, the information analysis instruction can be generated by filling each text key field into a corresponding position in the preset template according to the field data attribute of each text key field.

[0118] For example, when the data extraction dimension is a news title, a news core paragraph and a key entity field, the preset template can be “classification task: determine the type information of the following financial information (type information: first type, second type, third type and fourth type); news title: [title]; news content: [core paragraph]; key entity: [entity list]”.

[0119] S403, generate candidate recommended information corresponding to the to-be-displayed text according to the information classification model and the information analysis instruction.

[0120] Optionally, the information analysis instruction can be input into the information classification model to perform semantic analysis on the information in the information analysis instruction, so as to obtain the type information of the to-be-displayed text. For example, to improve the text processing efficiency, a plurality of information analysis instructions can be input into the information classification model at the same time. Specifically, a batch packaging principle can be used to splice a preset number of single information analysis instructions into a batch input instruction, and a separator (such as “###”) is used to distinguish between each information analysis instruction. At this time, the number of text units (tokens) in the batch input instruction needs to be controlled to be less than the inputtable number of the information classification model.

[0121] Optionally, for each to-be-displayed text associated with the information display page, type information of the to-be-displayed text can be added to a preset position in the to-be-displayed text, so as to form candidate recommended information corresponding to the to-be-displayed text. Alternatively, the type information of the to-be-displayed text and the to-be-displayed text can be directly spliced to obtain the candidate recommended information.

[0122] In the embodiment of the application, by using the information classification model to analyze the to-be-displayed text, the candidate recommended information containing the type information is obtained, so that the reliability of the candidate recommended information is ensured.

[0123] Figure 5 For another flowchart of the information recommendation method in the embodiment, on the basis of the above embodiment, an optional example of the information recommendation method is provided in the embodiment. In combination with Figure 5 , the implementation process is as follows:

[0124] S501, using a deep interest evolution network (DIEN), historical behavior data of a target user on an information display page is processed to obtain short-term interest change and long-term interest change of the target user on each type of information in the financial field.

[0125] The short-term interest change includes a short-term interest type and a first interest change trend, and the long-term interest change includes a long-term interest type and a second interest change trend.

[0126] S502, user attribute information and historical behavior data of the target user are extracted to obtain user attribute features and user behavior features.

[0127] S503, using the user attribute features and the user behavior features, each leaf node in a user behavior feature tree is traversed to obtain a hit target leaf node.

[0128] S504, feature information associated with the target leaf node is processed to obtain interest statistics of the target user on each type of information in the financial field.

[0129] The user behavior feature tree is constructed according to sample behavior data and sample attribute information.

[0130] S505, according to information acquisition frequency of the target user on the information display page, an interest adjustment weight is determined.

[0131] S506, according to the interest adjustment weight, interest change and interest statistics of the target user on each type of information in the financial field are processed to obtain interest scores of each type of information.

[0132] Optionally, the dynamic interest probability corresponding to each type of information is determined according to the change of the interest of the target user in each type of information in the financial field, the static interest probability corresponding to each type of information is determined according to the interest statistics of the target user in each type of information in the financial field, and the dynamic interest probability and the static interest probability corresponding to each type of information are processed according to the interest adjustment weight to obtain the interest score of each type of information.

[0133] In S507, the target recommendation information is determined from each candidate recommendation information according to the interest score of each type of information.

[0134] Optionally, for each to-be-displayed text associated with the information display page, the to-be-displayed text is subjected to semantic analysis processing to obtain a text key field corresponding to the to-be-displayed text; the text key field is input into a preset template associated with the information classification model to obtain an information analysis instruction; and the candidate recommendation information corresponding to the to-be-displayed text is generated according to the information classification model and the information analysis instruction.

[0135] The specific process of S501-S507 can be referred to the description of the method embodiments, and the implementation principle and technical effects are similar, which will not be described here.

[0136] It should be understood that, although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise stated herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0137] Based on the same inventive concept, the embodiments of the present application also provide an information recommendation device for implementing the information recommendation method described above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, and therefore the specific limitations in one or more information recommendation device embodiments provided below can be referred to the limitations of the information recommendation method described above, which will not be described here.

[0138] In one exemplary embodiment, as shown in Figure 6 An information recommendation device 1 is provided, which includes a first determination model 10, a second determination model 20 and an information feedback module 30, wherein:

[0139] The first determination model 10 is configured to determine a change of interest of the target user in each type of information in the financial field according to historical behavior data of the target user on the information display page.

[0140] The second determination model 20 is configured to determine a statistical interest of the target user in each type of information in the financial field according to user attribute information and historical behavior data of the target user and a hit on each leaf node in a user behavior feature tree; the user behavior feature tree is constructed according to sample behavior data and sample attribute information.

[0141] The information feedback module 30 is configured to determine target recommended information from each candidate recommended information according to the change of interest and the statistical interest of the target user in each type of information in the financial field, and recommend the target recommended information to the target user through the information display page.

[0142] In an exemplary embodiment, the change of interest includes a short-term change of interest and a long-term change of interest; the first determination model 10 is specifically configured to:

[0143] The historical behavior data is processed by using a deep interest evolution network DIEN to obtain the short-term change of interest and the long-term change of interest of the target user in each type of information in the financial field; the short-term change of interest includes a short-term interest type and a first interest change trend; the long-term change of interest includes a long-term interest type and a second interest change trend.

[0144] In an exemplary embodiment, the second determination model 20 is specifically configured to:

[0145] The user attribute information and the historical behavior data of the target user are extracted to obtain user attribute features and user behavior features; the user attribute features and the user behavior features are used to traverse each leaf node in the user behavior feature tree to obtain a target leaf node; and the feature information associated with the target leaf node is processed to obtain the statistical interest of the target user in each type of information in the financial field.

[0146] In an exemplary embodiment, the information feedback module 30 is specifically configured to:

[0147] The interest adjustment weight is determined according to the information acquisition frequency of the target user on the information display page; the change of interest and the statistical interest of the target user in each type of information in the financial field are processed according to the interest adjustment weight to obtain an interest score of each type of information; and the target recommended information is determined from each candidate recommended information according to the interest score of each type of information.

[0148] In an exemplary embodiment, the information feedback module 30 is further configured to:

[0149] According to the interest change of the target user to each type of information in the financial field, the dynamic interest probability corresponding to each type of information is determined; according to the interest statistical situation of the target user to each type of information in the financial field, the static interest probability corresponding to each type of information is determined; the dynamic interest probability and the static interest probability corresponding to each type of information are processed according to the interest adjustment weight, and the interest score of each type of information is obtained.

[0150] In an exemplary embodiment, the information recommendation apparatus 1 further comprises an information processing module, wherein the information processing module is specifically configured to:

[0151] For each to-be-displayed text associated with the information display page, the to-be-displayed text is subjected to semantic analysis processing to obtain a text key field corresponding to the to-be-displayed text; the text key field is input into a preset template associated with the information classification model to obtain an information analysis instruction; and the candidate recommended information corresponding to the to-be-displayed text is generated according to the information classification model and the information analysis instruction.

[0152] Each module in the information recommendation apparatus described above can be realized wholly or partially by software, hardware, and a combination thereof. Each module described above can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform the operations corresponding to each module.

[0153] In an exemplary embodiment, a computer device is provided, which can be a server, and an internal structure diagram thereof can be as shown in Figure 7 The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is configured to store user behavior data. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through a network connection. The computer program is executed by the processor to implement an information recommendation method.

[0154] Those skilled in the art can understand that Figure 7The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0155] In one embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor implementing the steps in the above method embodiments when executing the computer program.

[0156] In one embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program implements the steps in the above method embodiments when executed by a processor.

[0157] In one embodiment, a computer program product is provided, including a computer program, and the computer program implements the steps in the above method embodiments when executed by a processor.

[0158] It should be noted that the user information (including but not limited to user attribute information, etc.) and data (including but not limited to user behavior data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0159] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. The volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0160] The technical features of the above embodiments can be combined in any manner. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.

[0161] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. An information recommendation method characterized by comprising: The method comprises: determining a change in interest of a target user in each type of information in a financial field according to historical behavior data of the target user on an information display page; determining an interest statistical situation of the target user in each type of information in the financial field according to user attribute information of the target user and a hit situation of each leaf node on a user behavior feature tree according to the historical behavior data, wherein the user behavior feature tree is constructed according to sample behavior data and sample attribute information; determining target recommended information from each candidate recommended information according to the change in interest and the interest statistical situation of the target user in each type of information in the financial field, and recommending the target recommended information to the target user through the information display page.

2. The method of claim 1, wherein, The change in interest comprises a short-term change in interest and a long-term change in interest. The determining of the change in interest of the target user in each type of information in the financial field according to the historical behavior data of the target user on the information display page comprises: processing the historical behavior data by using a deep interest evolution network (DIEN) to obtain the short-term change in interest and the long-term change in interest of the target user in each type of information in the financial field, wherein the short-term change in interest comprises a short-term interest type and a first interest change trend, and the long-term change in interest comprises a long-term interest type and a second interest change trend.

3. The method of claim 1, wherein, The determining of the interest statistical situation of the target user in each type of information in the financial field according to the user attribute information of the target user and the hit situation of each leaf node on the user behavior feature tree comprises: extracting features from the user attribute information of the target user and the historical behavior data to obtain user attribute features and user behavior features; traversing each leaf node in the user behavior feature tree by using the user attribute features and the user behavior features to obtain a target leaf node that is hit; processing feature information associated with the target leaf node to obtain the interest statistical situation of the target user in each type of information in the financial field.

4. The method of claim 1, wherein, The determining of the target recommended information from each candidate recommended information according to the change in interest and the interest statistical situation of the target user in each type of information in the financial field comprises: determining an interest adjustment weight according to an information acquisition frequency of the target user on the information display page; processing the change in interest and the interest statistical situation of the target user in each type of information in the financial field according to the interest adjustment weight to obtain an interest score of each type of information; determining the target recommended information from each candidate recommended information according to the interest score of each type of information.

5. The method of claim 4, wherein, The processing of the change in interest and the interest statistical situation of the target user in each type of information in the financial field according to the interest adjustment weight to obtain the interest score of each type of information comprises: determining a dynamic interest probability corresponding to each type of information according to the change in interest of the target user in each type of information in the financial field. According to the interest statistical situation of the target user on each type of information in the financial field, a static interest probability corresponding to each type of information is determined; According to the interest adjustment weight, the dynamic interest probability and the static interest probability corresponding to each type of information are processed to obtain an interest score of each type of information.

6. The method of claim 1, wherein, The method further comprises: For each to-be-displayed text associated with the information display page, the to-be-displayed text is subjected to semantic analysis processing to obtain a text key field corresponding to the to-be-displayed text; The text key field is input into a preset template associated with an information classification model to obtain an information analysis instruction; According to the information classification model and the information analysis instruction, candidate recommended information corresponding to the to-be-displayed text is generated.

7. An information recommendation device characterized by comprising: The device comprises: A first determination model configured to determine, according to historical behavior data of a target user on an information display page, an interest change situation of the target user on each type of information in a financial field; A second determination model configured to determine, according to user attribute information of the target user and a hit situation of the historical behavior data on each leaf node of a user behavior feature tree, an interest statistical situation of the target user on each type of information in the financial field, wherein the user behavior feature tree is constructed according to sample behavior data and sample attribute information; An information feedback module configured to determine, according to the interest change situation and the interest statistical situation of the target user on each type of information in the financial field, a target recommended information from each candidate recommended information, and recommend the target recommended information to the target user through the information display page.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

Images