Information mining method, device and equipment for easily-visited user, medium and product

Abstract

The invention discloses an information mining method, device and equipment for an easily-visited user, a medium and a product, and the method comprises the steps: recognizing the easily-visited user based on user data; constructing a graph structure according to the product themes concerned by the easily visited users, the product themes of operators and the visited recommendation frequencies and visited frequencies of different product themes; and based on the information in the graph structure, according to a preset graph convolutional neural network model, obtaining the visited recommendation frequency and visited frequency of the product theme associated with the visited user at the future moment. According to the invention, the accuracy of information mining of easily-visited users can be improved.

Description

Technical Field

[0001] The present application relates to the field of data mining technology, and in particular to a method, device, equipment, medium and product for mining information of susceptible users. Background Art

[0002] With the rapid development of information technology, operators have accumulated and accumulated a vast amount of data. This surge in data contains a wealth of important information, but currently there is a lack of effective utilization of this data. Existing technologies utilize the data mining capabilities of machine learning algorithms to achieve simple identification of users who are likely to be interviewed and predict the probability of successful interviews. However, existing technologies simply predict interview probability based on historical user behavior data without considering the user's perceived fairness after the interview. Consequently, the accuracy of current interviewed user information mining is low. Summary of the Invention

[0003] To solve the above technical problems, the present application provides a method, device, equipment, medium and product for mining information of easily interviewed users, which can improve the accuracy of mining information of easily interviewed users.

[0004] This embodiment of the present application provides a method for mining information of easily interviewed users, including:

[0005] Identify users who are likely to be interviewed based on user data;

[0006] Constructing a graph structure based on the product themes that are likely to be of interest to the interviewed users, the product themes of the operators, and the recommendation frequency and interview frequency of different product themes;

[0007] Based on the information in the graph structure and according to a preset graph convolutional neural network model, the recommendation frequency and the interview frequency of the product topics that are likely to be associated with the interviewed users at a future moment are obtained.

[0008] As an improvement to the above solution, the graph structure is constructed based on the product themes that are likely to be of interest to the interviewed users, the product themes of the operators, and the interview recommendation frequency and interview frequency of different product themes, including:

[0009] For the user who is easily interviewed, obtain the product themes that the user is interested in, the product themes of the operator, the frequency of the operator's interview recommendations for different product themes, and the frequency of the user's positive emotional response to the product themes;

[0010] Take users, product themes, and operators as nodes, use the interviewed recommendation frequency as the edge between the operator node and the product theme node, use the interviewed frequency as the edge between the user node and the product theme node, and determine the edge between the user node and the operator node based on the interviewed frequency and the interviewed recommendation frequency corresponding to all product themes associated with the user, to obtain a graph structure.

[0011] As an improvement to the above solution, determining the edge between the user node and the operator node based on the visit frequency and the visit recommendation frequency corresponding to all product topics associated with the user includes:

[0012] For each user node, determine whether the interviewed frequency corresponding to all product topics associated with the user and the interviewed recommendation frequency have overlapping frequency intervals;

[0013] If yes, the overlapping frequency interval of the interviewed frequency and the interviewed recommendation frequency corresponding to all product themes associated with the user is used as the edge between the user node and the operator node;

[0014] If not, a frequency interval is determined according to the mean, maximum, and minimum values of the interview frequencies corresponding to all product topics associated with the user, and the frequency interval is used as the edge between the user node and the operator node.

[0015] As an improvement to the above solution, the method of obtaining the product themes that users are interested in, the product themes of operators, the frequency of operators' recommendations for different product themes, and the frequency of product themes for which users' sentiment after the interview is positive includes:

[0016] Extract product topics that users are interested in based on their product reviews;

[0017] Extract the operator's product theme based on the operator's product description text;

[0018] Based on the product themes of the operator, determine the number of users corresponding to each product theme according to the similarity between each product theme and the historical product acceptance of each user; obtain the historical frequency of recommendation received by the user, calculate the average of the historical frequency of recommendation received by the number of users corresponding to each product theme, and obtain the frequency of recommendation received by each product theme;

[0019] Obtain the frequency of users' interviews on different products and the emotional trends of users after each interview, and take the products for which the emotional trends of the users after the interviews are positive as positive emotional products; determine several positive emotional products corresponding to each product theme based on the similarity between the product themes that the users are interested in and the positive emotional products corresponding to the users; calculate the average of the frequency of interviews for the several positive emotional products corresponding to each product theme, and obtain the frequency of interviews for product themes for which the emotional trends of the users after the interviews are positive.

[0020] As an improvement to the above solution, the method of identifying users who are susceptible to being interviewed based on user data includes:

[0021] Obtaining multi-dimensional user data of the operator, and filtering the multi-dimensional user data to obtain input features;

[0022] Based on the input features, a user identification model that is susceptible to being interviewed is constructed, and users that are susceptible to being interviewed are obtained according to the user identification model that is susceptible to being interviewed;

[0023] The characteristic data of the user who is easily interviewed is obtained, and a preset Gaussian mixture model is used to perform classification based on the characteristic data to obtain a category label of the user who is easily interviewed.

[0024] As an improvement to the above solution, based on the information in the graph structure and according to a preset graph convolutional neural network model, obtaining the recommended frequency and interview frequency of the product topics that are likely to be associated with the interviewed users at a future moment includes:

[0025] Partitioning the graph structure according to the distance between each product theme node and an adjacent product theme node in the graph structure to obtain a plurality of subgraph structures;

[0026] Extracting a feature matrix of product topics in the subgraph structure;

[0027] The node information in the subgraph structure and the feature matrix are input into a preset graph convolutional neural network model to predict the recommendation frequency and visit frequency of the product topics that are likely to be associated with the interviewed users in the future.

[0028] The present application also provides an information mining device for users who are susceptible to being interviewed, including:

[0029] A user identification module, used to identify susceptible users based on user data;

[0030] A graph structure construction module is used to construct a graph structure based on the product themes that are likely to be of interest to the interviewed users, the product themes of the operators, and the frequency of recommendation and interview of different product themes;

[0031] The model prediction module is used to obtain the recommendation frequency and interview frequency of the product topics that are likely to be associated with the interviewed users at a future moment based on the information in the graph structure and according to a preset graph convolutional neural network model.

[0032] An embodiment of the present application further provides a computer device, comprising a processor and a memory, wherein a computer program is stored in the memory and configured to be executed by the processor, and wherein when the processor executes the computer program, any of the above-described methods for mining information on susceptible users is implemented.

[0033] An embodiment of the present application further provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute any of the above-mentioned methods for mining information on easily interviewed users.

[0034] An embodiment of the present application further provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned methods for mining information on easily interviewed users.

[0035] Compared with the prior art, the beneficial effects of the information mining method, device, equipment, medium and product of the easy-to-interview user provided by the embodiment of the present application are: by identifying the easy-to-interview user, and constructing a graph structure according to the product themes that the easy-to-interview user is concerned about, the product themes of the operator, and the frequency of recommended visits and the frequency of visits of different product themes, and then based on the information in the graph structure, the frequency of recommended visits and the frequency of visits of product themes associated with the easy-to-interview user in the future are predicted according to the graph convolutional neural network model, which can fully mine the implicit relationship between users, products and operators, improve the prediction accuracy, and thus improve the accuracy of interviewed user information mining; by The product themes of merchants and users, the frequency of operators' recommendations for different product themes, and the frequency of product themes with positive emotional trends among users after the interviews are constructed. A graph structure is constructed to mine the frequency characteristics of the interviews among users, products, and operators that can maintain the positive emotional trends of users. Then, the graph convolutional neural network model is combined to obtain the predicted frequency of recommendation and frequency of interviews. This improves the defect of the existing technology that does not consider the fairness perceived by the interviewed users after the interview, and can enhance the user's product usage experience. In addition, by partitioning the graph structure, the relevance of products can be retained to the maximum extent, and the obtained feature information can be enhanced, thereby improving the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flow chart of a method for mining information of easily interviewed users provided in an embodiment of the present application;

[0037] Figure 2 is a schematic diagram of a graph structure provided in an embodiment of the present application;

[0038] Figure 3 This is a structural diagram of a graph convolutional neural network model provided in an embodiment of the present application;

[0039] Figure 4 This is a schematic diagram of the structure of an information mining device for easily interviewed users provided in an embodiment of the present application;

[0040] Figure 5 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0042] When mining information about users who are susceptible to being interviewed, existing technologies simply predict the probability of being interviewed based on historical user-side behavioral data, without considering the fairness perceived by the interviewed users after the interview. Fairness refers to the average interval between the number of times a user receives surveys, marketing calls, or text messages, and the average similarity of the content types of the surveys, marketing calls, or text messages received within a fixed period of time. In other words, users maintain a positive emotional trend after receiving the surveys, marketing calls, or text messages. Therefore, existing methods can easily cause users to develop negative emotional tendencies toward the subjects associated with the surveys, marketing calls, or text messages. For example, for example, "Promise a minimum spend of 80 yuan to obtain membership benefits." For operators, after using existing technologies to identify a user as susceptible to being interviewed, they often frequently push marketing calls such as "Promise a minimum spend of 80 yuan to obtain membership benefits" to the user in a short period of time. Because existing technologies do not consider access fairness when mining and predicting user interview probabilities, this can cause users to develop negative emotional tendencies, reduce user experience satisfaction, increase user complaints, and ultimately cause user churn. Therefore, in order to solve the problem of low accuracy in information mining of interviewed users in the prior art, an embodiment of the present application provides an information mining method for easily interviewed users, which can improve the accuracy of information mining and help mine users who are willing to accept survey calls.

[0043] See also Figure 1 , Figure 1 : This is a flow chart of a method for mining information about users who are likely to be interviewed, provided in an embodiment of the present application. The method for mining information about users who are likely to be interviewed comprises:

[0044] S1: Identify users who are susceptible to being interviewed based on user data;

[0045] S2: Constructing a graph structure based on the product themes that are likely to be of interest to the interviewed users, the product themes of the operators, and the recommendation frequency and interview frequency of different product themes;

[0046] S3: Based on the information in the graph structure and according to a preset graph convolutional neural network model, obtain the recommended frequency and the interviewed frequency of the product topics that are likely to be associated with the interviewed users at a future moment.

[0047] Specifically, the process begins by acquiring user data from mobile operators, such as basic user information and communication behavior data. Feature extraction is then performed on this data, which is then fed into a model for identifying users who are likely to be interviewed. The model outputs a list of users who are likely to be interviewed. The model can be trained using a classifier, such as a random forest, gradient boosting tree, or support vector machine.

[0048] After obtaining the easily interviewed users, we then use their user data and operator data to obtain the product themes that these easily interviewed users care about, the product themes extracted from the operator data, and the frequency of recommendation and visit for different product themes. Each product theme includes several products, specifically various call packages and data plans from operators. For a given user, the frequency of recommendation for a product theme is the frequency with which the operator recommends products of that product theme to the user, and the frequency of visit for a product theme is the frequency with which the operator recommends products of that product theme to the user. Furthermore, based on the product themes that users care about, the product themes of operators, and the frequency of recommendation and visit for different product themes, we construct a graph structure with operators, product themes, and users as nodes. This graph represents all features, thereby exploring the implicit relationships between users, products, and operators, providing a basis for subsequent accurate prediction of user visit frequencies.

[0049] Based on the feature information and node information in the graph structure, a preset graph convolutional neural network model is used for prediction to obtain the frequency of recommendation and frequency of product themes associated with the susceptible users in the future, so that the frequency of recommendation of products of the operator for the product themes preferred by the susceptible users in the future can be known. Furthermore, user surveys or marketing are conducted based on the information of the susceptible users and the recommendation frequency of their preferred products obtained through mining. The embodiment of the present application takes into account the fairness of user access, can effectively mine users who are willing to accept survey calls, improve the accuracy of mining information of susceptible users, avoid users from having negative emotions, and improve customer experience.

[0050] As one of the optional embodiments, identifying the susceptible users based on the user data includes:

[0051] Obtaining multi-dimensional user data of the operator, and filtering the multi-dimensional user data to obtain input features;

[0052] Based on the input features, a user identification model that is susceptible to being interviewed is constructed, and users that are susceptible to being interviewed are obtained according to the user identification model that is susceptible to being interviewed;

[0053] The characteristic data of the user who is easily interviewed is obtained, and a preset Gaussian mixture model is used to perform classification based on the characteristic data to obtain a category label of the user who is easily interviewed.

[0054] Specifically, the operator's multi-dimensional user data is obtained, and the multi-dimensional user data includes user basic attribute data, package benefit usage data, answering preference type data, abnormal situation monitoring data and other special attribute data.

[0055] For example, the basic attribute data of the user includes: user ID, mobile phone number, age, gender, network age, star rating, ARPU, tariff price, number of days of network access, etc.;

[0056] Package benefit usage data includes: total voice time of the package + voice package, voice utilization rate for the current month, MOU for the current month, available data traffic for the current month, data utilization rate for the current month, percentage of days with voice usage, percentage of days with data usage, number of equity products subscribed, and utilization rate of equity products.

[0057] Answering preference data includes: number of called calls, duration of called calls, call connection rate, number of calls from strangers (outside of social circle), duration of stranger calls, connection rate of stranger calls, number of calls from non-personal numbers (such as 95 / 96, 400 numbers), duration of calls from non-personal numbers, connection rate of calls from non-personal numbers, etc.;

[0058] Abnormal data monitoring includes: this month's ARPU / average ARPU of the past three months, this month's traffic usage / average traffic usage of the past three months, average monthly voice overage fees of the past three months / (ARPU + this month's package fee), whether there have been any package fee changes in the past three months, whether there have been any value-added service cancellations or non-renewals in the past three months, whether there has been any forward traffic suppression in the past three months, whether there has been any backward traffic suppression in the past three months, the number of months remaining in terminal contracts, and the number of equity package subscriptions.

[0059] Other special attribute data include: whether the user is silent for a long time, whether the customer is a group customer, whether the customer has a low-consumption insurance number, whether the customer has exceeded the high-value and high-frequency traffic limit, whether the customer has double reduction, whether the customer has improved the main card of another network, whether the customer has a single mobile phone without integration, whether the customer is on the blacklist, whether the customer is a home network customer, etc.

[0060] Furthermore, modeling features are screened from multi-dimensional user data. Feature selection removes redundant and irrelevant features from the original dataset, retaining valuable features, known as modeling features. Specifically, modeling features are obtained through a variety of pre-defined feature engineering methods, including but not limited to MI (mutual information), MIC (maximum mutual information coefficient), and the Pearson correlation coefficient. Mutual information is a measure of interdependence between variables, representing the amount of information shared between features and targets, and capturing both nonlinear and linear relationships. The maximum mutual information coefficient measures the degree of correlation between two variables. It discretizes the relationship between the two variables in a two-dimensional space and uses a scatter plot to represent it. The two-dimensional space is divided into a certain number of intervals, and the location of the current scatter plot in each square is then checked to calculate the joint probability. This solves the difficulty of determining joint probability in mutual information and maximizes the strength of detecting any relationship. The Pearson correlation coefficient measures the relationship between two variables. By calculating the Pearson correlation coefficient, the degree of correlation between the two variables can be determined, allowing for the elimination of one of the two highly correlated variables as needed to achieve redundancy removal. By combining multiple feature engineering methods, the most valuable features for model entry can be screened out quickly and efficiently.

[0061] Furthermore, based on the input features obtained through screening, a user identification model for easy interviews is constructed. Specifically, the XGBoost classifier is selected as the user identification model for easy interviews. XGBoost is an optimized distributed gradient boosting library designed to be efficient, flexible and portable. The idea is to integrate many weak classifiers together to form a strong classifier, and introduce a regularization term in the loss function to control the complexity of the model and prevent the model from overfitting. In addition, XGBoost can automatically learn the split direction and support column sampling. After each iteration, the model multiplies the weight of the leaf node by a learning rate, which can weaken the influence of each tree and allow for more learning space later. When splitting a node, XGBoost needs to calculate the gain of each feature, and finally select the feature with the largest gain for splitting, ensuring that the gain calculation of each feature can be performed in multiple threads.

[0062] Specifically, XGBoost is an additive model composed of k base models. Assume that the tree model to be trained in the tth iteration is f t (x), then:

[0063]

[0064] in represents the prediction result of sample i after the tth iteration, f t (xi ) represents the model of the t-th tree. According to the second-order Taylor expansion, the objective function of the t-th tree model can be expressed as:

[0065]

[0066] where g i represents the first-order derivative, h i represents the second-order derivative.

[0067] Put f t (x i ) is defined as w q (x i ), Q represents the leaf node where sample i is located, w represents the weight of the leaf node, and the regularization term is introduced, then the expression can be simplified to:

[0068]

[0069] make Simplifying, we can get:

[0070]

[0071] Derivative the above formula and set it to 0 to obtain the optimal weight value of the leaf node:

[0072]

[0073] Then the optimal solution of the objective function is:

[0074]

[0075] Then, the feature nodes are divided based on information gain, and the expression is as follows:

[0076]

[0077] The first part of the expression is the score of the new left cotyledon (i.e., the objective function of the left cotyledon after the node undergoes feature splitting); the second part is the score of the new right cotyledon (i.e., the objective function of the right cotyledon after the node undergoes feature splitting); the third part is the score of the original leaf (i.e., the objective function of the node before the feature split); and the fourth part is the regularization coefficient for the newly added leaf. The model can determine whether the information gain after splitting the node (the first part + the second part) is greater than that without the split, and consider whether the model is too complex (if the increased score is less than the regularization term, the node is not split again).

[0078] Furthermore, during the construction of the susceptible user identification model, the input features are divided into training data and verification data. The training data is used for model training to initially generate the susceptible user identification model. Furthermore, the verification data is input into the initially generated susceptible user identification model for model verification. Specifically, the model's effectiveness is comprehensively evaluated by calculating precision and recall rates. Based on the model evaluation results, the model parameters are iteratively optimized until the model evaluation results reach the preset indicator values, thereby obtaining the final susceptible user identification model. The final susceptible user identification model is used to output a list of susceptible users.

[0079] Furthermore, after identifying the list of users who are easily interviewed, the embodiment of the present application also performs further data mining on the users who are easily interviewed. Specifically, based on the characteristic data of the users who are easily interviewed, the Gaussian mixture model GMM is used to classify the users who are easily interviewed. GMM (Gaussian Mixture Model) is a probability model that can be used for segmentation. It is an extension of a single Gaussian probability density function. GMM can smoothly approximate the density distribution of any shape. Similar to clustering, each Gaussian model can be regarded as a category according to the different parameters of the Gaussian probability density function. By inputting a sample x, its value can be calculated by the Gaussian probability density function, and then a threshold is used to determine whether the sample belongs to the Gaussian model. GMM is suitable for multi-category division and can be applied to complex object modeling.

[0080] Among them, the characteristic data of users who are easily interviewed can be selected according to actual needs. For example, if it is necessary to focus on the user's product ordering intention, multiple data representing the user's product ordering dimensions are selected as characteristic data. At this time, the trained Gaussian mixture model will output the product ordering intention label of each user who is easily interviewed, which is divided into low ordering intention, medium ordering intention and high ordering intention; if it is necessary to focus on the user's consumption ability, multiple data representing the consumption ability dimension are selected as characteristic data. At this time, the trained Gaussian mixture model will output the consumption ability label of each user who is easily interviewed, which is divided into low consumption ability, medium consumption ability and high consumption ability.

[0081] By clustering the Gaussian mixture model (GMM), we can obtain category labels for users who are likely to be interviewed, and achieve refined classification of users. This can significantly improve the pertinence and effectiveness of user contact, conduct precise outbound surveys, and help improve user experience.

[0082] In another optional embodiment, the identifying of easily interviewed users based on user data specifically includes: obtaining user data, which includes basic user information, call records, and bill records; performing feature extraction on the user data to obtain feature data, including call features, bill features, customer service interaction features, device and network features, etc.; further, inputting the feature data into a preset neural network model to obtain an easily interviewed score for each user output by the neural network model; and screening out users whose easily interviewed scores are greater than a preset threshold to obtain easily interviewed users.

[0083] Furthermore, after identifying the easily interviewed users, the embodiment of the present application conducts user information mining on the easily interviewed users, specifically constructing a graph structure based on the product themes that the easily interviewed users are interested in, the product themes of the operators, and the recommended frequency and frequency of interviews of different product themes. Then, based on the information in the graph structure and according to the preset graph convolutional neural network model, the recommended frequency and frequency of interviews of the product themes associated with the easily interviewed users at future moments are obtained, thereby achieving more refined easily interviewed user information mining, which is conducive to improving user experience.

[0084] As one of the optional embodiments, constructing a graph structure based on the product themes that are likely to be of interest to the interviewed users, the product themes of the operators, and the interview recommendation frequencies and interview frequencies of different product themes includes:

[0085] For the user who is easily interviewed, obtain the product themes that the user is interested in, the product themes of the operator, the frequency of the operator's interview recommendations for different product themes, and the frequency of the user's positive emotional response to the product themes;

[0086] Take users, product themes, and operators as nodes, use the interviewed recommendation frequency as the edge between the operator node and the product theme node, use the interviewed frequency as the edge between the user node and the product theme node, and determine the edge between the user node and the operator node based on the interviewed frequency and the interviewed recommendation frequency corresponding to all product themes associated with the user, to obtain a graph structure.

[0087] The acquisition of product themes that users are interested in, product themes that operators are interested in, the frequency of operators recommending different product themes, and the frequency of product themes for which users have a positive sentiment after being interviewed include:

[0088] Extract product topics that users are interested in based on their product reviews;

[0089] Extract the operator's product theme based on the operator's product description text;

[0090] Based on the product themes of the operator, determine the number of users corresponding to each product theme according to the similarity between each product theme and the historical product acceptance of each user; obtain the historical frequency of recommendation received by the user, calculate the average of the historical frequency of recommendation received by the number of users corresponding to each product theme, and obtain the frequency of recommendation received by each product theme;

[0091] Obtain the frequency of users' interviews on different products and the emotional trends of users after each interview, and take the products for which the emotional trends of the users after the interviews are positive as positive emotional products; determine several positive emotional products corresponding to each product theme based on the similarity between the product themes that the users are interested in and the positive emotional products corresponding to the users; calculate the average of the frequency of interviews for the several positive emotional products corresponding to each product theme, and obtain the frequency of interviews for product themes for which the emotional trends of the users after the interviews are positive.

[0092] Specifically, for the identified susceptible users, product reviews are obtained and preprocessed. Based on the star rating of each product review, the product reviews are divided into positive and negative. For example, five-star reviews are considered positive, while one-star reviews are considered negative. An equal number of positive and negative reviews are then input into a pre-defined structural topic model (STM) to extract product topics. The star rating is also used as a covariate in the STM to reflect the sentiment trends of each product topic. Finally, the optimal number of topics and iterations of the STM are set, and the STM is used to extract product topics that users are interested in, reflecting their sentiment trends. Sentiment trends can be positive or negative. For example, users' sentiment trends for product topics corresponding to five-star reviews are positive, while those for one- or two-star reviews are negative. The product topics that users are interested in are semantic clusters aggregated from the reviews using the structural topic model. These clusters represent product attributes that users are interested in, such as network stability, package pricing, and customer service efficiency.

[0093] Furthermore, based on the products provided by operators for different users, such as low-priced packages and campus-exclusive traffic for students, international roaming packages and exclusive customer service for business groups, etc., the product description texts of these products are obtained and preprocessed, and the preprocessed product description texts are input into the preset structural topic model STM for topic extraction. The optimal number of topics and number of iterations of STM are set, and the product themes of the operators are extracted through STM.

[0094] Furthermore, the frequency of product theme recommendations is obtained. First, the user's historical acceptance of products after the interview is obtained; the similarity between each product theme of the operator and the historical acceptance of products of each user is calculated; for each product theme, several users whose similarity meets the preset conditions are selected to obtain the number of users corresponding to the product theme. Specifically, users corresponding to similarities greater than a preset threshold can be selected, or the top several users can be selected based on the similarity size; then, the user's historical frequency of recommendation for different products is obtained, that is, the frequency at which users actually accepted recommendations after the operator recommended products to them in the past; the average of the historical frequency of recommendation for several users corresponding to each product theme is calculated to obtain the frequency of recommendation for each product theme.

[0095] Furthermore, the recommendation frequency of product themes whose emotional tendency after the user is interviewed is positive is obtained. First, the frequency of user interviews for different products is obtained; the emotional tendency of the user after the interview for different products is analyzed, and the emotional tendency is divided into positive emotion and negative emotion. Specifically, the emotional tendency of the user after the interview can be obtained by collecting the user interview duration data or the voice content of the user during the call. For example, the emotional tendency of the user after the interview is determined based on the difference between the user interview duration and other call durations, or the difference between the user interview duration and the average interview duration. If the difference is greater than or equal to 0, it indicates a positive emotional tendency. If the difference is less than 0, it indicates a negative emotional tendency. If the voice content is used to determine the emotional tendency, the emotion recognition model can be used to analyze the voice text to obtain the emotional tendency of the user after the interview. Then, products whose emotional tendency after the user is interviewed is positive are screened out to obtain positive emotional products; the similarity between the product theme that the user is concerned about and the various products whose emotional tendency after the user is interviewed is positive is calculated, and several products whose similarity meets the preset conditions are selected to obtain several products corresponding to the product theme. Specifically, products corresponding to similarities greater than a preset threshold can be selected, or the top several products can be selected by sorting according to the size of the similarity; the average of the interview frequencies of several products that meet the preset conditions is calculated to obtain the interview frequency of product themes whose emotional tendency after the user is interviewed is positive.

[0096] Among them, by matching the product themes that users are interested in with the same or similar product themes among the operator's product themes, the product themes of the two are unified to facilitate the subsequent construction of the graph structure.

[0097] Furthermore, with users, product themes, and operators as nodes, the frequency of interviews and recommendations is used as the edge between the operator node and the product theme node, and the frequency of interviews is used as the edge between the user node and the product theme node. The edge between the user node and the operator node is determined based on the frequency of interviews and the frequency of interviews and recommendations corresponding to all product themes associated with the user, and a graph structure is constructed, such as Figure 2 shown.

[0098] Among them, operators correspond to a variety of product themes corresponding to different products. The edge between each product theme and the operator is the frequency of the operator's recommendation for the product theme. Each product theme can be applicable to multiple users, such as Figure 2 In the example, product topic 1 corresponds to multiple users to be interviewed. The edge between the product topic and each user represents the frequency of the user's positive sentiment towards the product topic. Each product topic corresponds to multiple product sets, consisting of multiple products with the same or similar attribute themes.

[0099] In actual application, there is also an association relationship between users and operators, that is, the edge between user nodes and operator nodes ( Figure 2 Not marked in the table), which is determined by the frequency of responses and recommended responses corresponding to all product topics associated with the user.

[0100] As one of the optional embodiments, determining the edge between the user node and the operator node according to the visit frequency and the visit recommendation frequency corresponding to all product themes associated with the user includes:

[0101] For each user node, determine whether the interviewed frequency corresponding to all product topics associated with the user and the interviewed recommendation frequency have overlapping frequency intervals;

[0102] If yes, the overlapping frequency interval of the interviewed frequency and the interviewed recommendation frequency corresponding to all product themes associated with the user is used as the edge between the user node and the operator node;

[0103] If not, a frequency interval is determined according to the mean, maximum, and minimum values of the interview frequencies corresponding to all product topics associated with the user, and the frequency interval is used as the edge between the user node and the operator node.

[0104] Specifically, the overlap between the visit frequency of all product topics associated with the user and the visit frequency of the operator's recommendations for these product topics serves as a quantitative representation of the relationship between the operator and the user. If there is no overlap between the visit frequency and the visit frequency of all product topics associated with the user, the frequency interval is determined by the mean, minimum, and maximum visit frequencies of all product topics associated with the user, and this frequency interval is used as the edge between the user node and the operator node.

[0105] In a specific example, the mean is adjusted by the average of the differences between the mean and the maximum and minimum values, and the calculated average value is subtracted from the mean as the starting value of the frequency interval, and the calculated average value is added to the mean as the ending value of the frequency interval. In other words, determining the frequency interval based on the mean, maximum, and minimum values of the survey frequencies corresponding to all product topics associated with the user includes:

[0106] Calculate the mean μ and maximum f of the interview frequency corresponding to all product topics associated with the user max With the minimum value f min , and get the frequency interval: The embodiments of the present application are simple to calculate, can quickly generate results, save time and resources, are applicable to various data types, and are highly interpretable.

[0107] In another specific example, according to the mean μ and maximum f of the interview frequencies corresponding to all product topics associated with the user, max With the minimum value f min , calculate the frequency range: [μ-k(μ-f min ), μ+k(f max -μ)], where k is the adjustment coefficient.

[0108] Furthermore, in the distribution of all product theme nodes in the graph structure, the distance between product theme nodes is set according to the similarity of the user types of the user nodes associated with different product theme nodes, so that the user types of the user nodes associated with the product theme nodes and the user nodes associated with the spatially adjacent product theme nodes have a high similarity. For example, in the graph structure, the adjacent product theme nodes of product theme 1 in spatial distribution are product theme 2 and product theme 6. If the calculated similarity of the user type of the user associated with product theme 2 and the user type of the user associated with product theme 1 is greater than that of product theme 6, then in the graph structure, the distance between product theme 1 and product theme 2 is less than the distance between product theme 1 and product theme 6. By setting the node distribution of the graph structure, the relationship between the nodes can be more intuitively observed, which improves the accuracy of subsequent user information mining.

[0109] Furthermore, the information in the graph structure is input into a pre-built graph convolutional neural network model to further mine and obtain information on susceptible users, which is then returned to the operator for investigation.

[0110] As one of the optional embodiments, based on the information in the graph structure and according to a preset graph convolutional neural network model, obtaining the recommended frequency and the interviewed frequency of the product topics that are likely to be associated with the interviewed users at a future moment includes:

[0111] Partitioning the graph structure according to the distance between each product theme node and an adjacent product theme node in the graph structure to obtain a plurality of subgraph structures;

[0112] Extracting a feature matrix of product topics in the subgraph structure;

[0113] The node information in the subgraph structure and the feature matrix are input into a preset graph convolutional neural network model to predict the recommendation frequency and visit frequency of the product topics that are likely to be associated with the interviewed users in the future.

[0114] Each of the subgraph structures includes an operator node, at least one product theme node and several user nodes.

[0115] Specifically, obtain real-time data from users and operators and map the real-time data to the graph structure G d = {V, E, A}, where V∈R n Represents the node set of the graph, n is the number of nodes in the graph, E represents the edge set of the graph, A∈R n×n Represents the adjacency matrix of a graph.

[0116] For each operator's graph structure, the distance between each product theme node and the adjacent product theme node is determined based on the similarity between the user types of users associated with the two adjacent product theme nodes; the product theme nodes whose distance between nodes is less than the preset distance threshold are divided into a partition, and several partitions are obtained. Each partition is a subgraph structure, and the subgraph structure is represented as G ′ d (x) = {V′, E′, A′}, where V′∈R n′ Represents the node set of the subgraph structure, n′ is the number of nodes in the subgraph structure, E′ represents the edge set of the subgraph structure, A′∈R n′×n′ The adjacency matrix represents the subgraph structure. By partitioning the graph structure and then processing the subgraph structure, the product relevance can be preserved to the greatest extent, the obtained feature information can be enhanced, and the prediction accuracy can be improved.

[0117] Furthermore, according to the out-degree matrix and adjacency matrix of the subgraph structure, the product topic matrix T = D′-A′ of each subgraph structure is calculated, where T∈R n′×n′ , D′∈R n′×n′ , G ′ d The out-degree matrix D′ of (x) is a diagonal matrix, and the corresponding diagonal values represent the out-degree of the node. Then, through the graph Fourier transform, the characteristic matrix of the product topic matrix T is extracted to obtain the characteristic matrix of the product topic in each subgraph structure.

[0118] Furthermore, the node information and feature matrix in the subgraph structure are input into the pre-built graph convolutional neural network model, where the node information includes the user information, operator information and user information of the adjacent product theme nodes on each node in the subgraph structure. The structure of the graph convolutional neural network model is as follows: Figure 3 As shown in the figure, the model consists of a graph convolutional neural network (GCN), a normalization layer (BatchNorm2D), an activation function (PReLU), a two-dimensional convolutional layer (Conv2D), a normalization layer (BatchNorm2D), and a series of residual-connected CNN convolutional modules. After passing through the input layer, the model's input data is convolved by the graph convolutional neural network (GCN). It is then normalized using BatchNorm2D. After activation with the PReLU activation function, it undergoes a two-dimensional convolution by Conv2D. It is then normalized using BatchNorm2D and passed to a convolutional module consisting of a series of residual-connected CNNs. These convolutional modules fully exploit the spatiotemporal variations of features to predict the frequency of recommendation and visit of product themes that are likely to be associated with users in the future. Furthermore, by combining the graph structure with the predicted recommendation frequency and interview frequency, we can obtain the operator's recommendation frequency for products corresponding to the product themes preferred by each likely user in the future. This allows for research and improves the existing technology's flaw of not considering the fairness perceived by the interviewed users after the interview. This can meet users' emotional needs for product usage, avoid user churn, and improve user stability and stickiness.

[0119] The embodiments of the present application identify users who are susceptible to being interviewed, and construct a graph structure based on the product themes that are of interest to the susceptible users, the product themes of the operators, and the frequency of recommendation and frequency of interviews for different product themes. Based on the information in the graph structure, the graph convolutional neural network model is used to predict the frequency of recommendation and frequency of interviews for product themes that are likely to be associated with the susceptible users in the future. This fully mines the implicit relationships between users, products, and operators, improves prediction accuracy, and thereby improves the accuracy of interviewed user information mining. A graph structure is constructed based on the product themes of operators and users, the frequency of recommendation by operators for different product themes, and the frequency of interviews for product themes with positive post-interview sentiment. This allows for the mining of frequency characteristics between users, products, and operators that can maintain a positive sentiment trend for users. The graph convolutional neural network model is then used to obtain predicted frequency of recommendation and frequency of interviews. This overcomes the drawback of existing technologies that fail to consider the fairness perceived by interviewed users after the interview, thereby improving the user's product usage experience. Furthermore, by partitioning the graph structure, the relevance of products can be maximized, and the obtained feature information can be enhanced, thereby improving prediction accuracy. The embodiments of the present application can fully mine the information of users who are easily interviewed, obtain a reasonable survey frequency based on the users' emotional feedback on the product, improve the quality of telecommunications services, and ensure user satisfaction.

[0120] Correspondingly, the present application also provides an information mining device for users who are easily interviewed, which can implement all the processes of the information mining method for users who are easily interviewed in the above embodiment.

[0121] See also Figure 4 , Figure 4 Schematic diagram of a device for mining information about users who are easily interviewed provided in an embodiment of the present application. The device for mining information about users who are easily interviewed comprises:

[0122] User identification module 401, used to identify susceptible users based on user data;

[0123] A graph structure building module 402 is configured to build a graph structure based on the product themes that are likely to be of interest to the interviewed users, the product themes of the operators, and the frequency of recommendation and interview of different product themes;

[0124] The model prediction module 403 is used to obtain the recommendation frequency and interview frequency of the product topics that are likely to be associated with the interviewed users at a future moment based on the information in the graph structure and according to a preset graph convolutional neural network model.

[0125] Preferably, the graph structure building module 402 is specifically used to:

[0126] For the user who is easily interviewed, obtain the product themes that the user is interested in, the product themes of the operator, the frequency of the operator's interview recommendations for different product themes, and the frequency of the user's positive emotional response to the product themes;

[0127] Take users, product themes, and operators as nodes, use the interviewed recommendation frequency as the edge between the operator node and the product theme node, use the interviewed frequency as the edge between the user node and the product theme node, and determine the edge between the user node and the operator node based on the interviewed frequency and the interviewed recommendation frequency corresponding to all product themes associated with the user, to obtain a graph structure.

[0128] Preferably, determining the edge between the user node and the operator node according to the visit frequency and the visit recommendation frequency corresponding to all product themes associated with the user includes:

[0129] For each user node, determine whether the interviewed frequency corresponding to all product topics associated with the user and the interviewed recommendation frequency have overlapping frequency intervals;

[0130] If yes, the overlapping frequency interval of the interviewed frequency and the interviewed recommendation frequency corresponding to all product themes associated with the user is used as the edge between the user node and the operator node;

[0131] If not, a frequency interval is determined according to the mean, maximum, and minimum values of the interview frequencies corresponding to all product topics associated with the user, and the frequency interval is used as the edge between the user node and the operator node.

[0132] Preferably, the obtaining of product themes of interest to users, product themes of operators, the frequency of operators' recommendations for different product themes, and the frequency of product themes for which users' sentiment after the interview is positive includes:

[0133] Extract product topics that users are interested in based on their product reviews;

[0134] Extract the operator's product theme based on the operator's product description text;

[0135] Based on the product themes of the operator, determine the number of users corresponding to each product theme according to the similarity between each product theme and the historical product acceptance of each user; obtain the historical frequency of recommendation received by the user, calculate the average of the historical frequency of recommendation received by the number of users corresponding to each product theme, and obtain the frequency of recommendation received by each product theme;

[0136] Obtain the frequency of users' interviews on different products and the emotional trends of users after each interview, and take the products for which the emotional trends of the users after the interviews are positive as positive emotional products; determine several positive emotional products corresponding to each product theme based on the similarity between the product themes that the users are interested in and the positive emotional products corresponding to the users; calculate the average of the frequency of interviews for the several positive emotional products corresponding to each product theme, and obtain the frequency of interviews for product themes for which the emotional trends of the users after the interviews are positive.

[0137] Preferably, the user identification module 401 is specifically used for:

[0138] Obtaining multi-dimensional user data of the operator, and filtering the multi-dimensional user data to obtain input features;

[0139] Based on the input features, a user identification model that is susceptible to being interviewed is constructed, and users that are susceptible to being interviewed are obtained according to the user identification model that is susceptible to being interviewed;

[0140] The characteristic data of the user who is easily interviewed is obtained, and a preset Gaussian mixture model is used to perform classification based on the characteristic data to obtain a category label of the user who is easily interviewed.

[0141] Preferably, the model prediction module 403 is specifically used to:

[0142] Partitioning the graph structure according to the distance between each product theme node and an adjacent product theme node in the graph structure to obtain a plurality of subgraph structures;

[0143] Extracting a feature matrix of product topics in the subgraph structure;

[0144] The node information in the subgraph structure and the feature matrix are input into a preset graph convolutional neural network model to predict the recommendation frequency and visit frequency of the product topics that are likely to be associated with the interviewed users in the future.

[0145] In specific implementation, the working principle, control process and technical effects achieved by the information mining device for easily interviewed users provided in the embodiment of the present application are the same as those of the information mining method for easily interviewed users in the above embodiment, and will not be repeated here.

[0146] See also Figure 5 , Figure 5is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. The computer device includes: a processor 501, a memory 502, and a computer program stored in the memory 502 and executable on the processor 501. When the processor 501 executes the computer program, the steps of the embodiment of the method for mining information on easily interviewed users are implemented. Alternatively, when the processor 501 executes the computer program, the functions of the modules / units in the above-mentioned apparatus embodiments are implemented.

[0147] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 502 and executed by the processor 501 to implement the present application. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program in the computer device.

[0148] The computer device may include, but is not limited to, a processor 501 and a memory 502. Those skilled in the art will appreciate that the schematic diagram is merely an example of a computer device and does not limit the computer device. The computer device may include more or fewer components than shown in the diagram, or may combine certain components or different components. For example, the computer device may also include input and output devices, network access devices, buses, and the like.

[0149] The processor 501 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor. The processor 501 is the control center of the computer device, connecting various parts of the entire computer device using various interfaces and lines.

[0150] The memory 502 can be used to store the computer programs and / or modules. The processor 501 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 502 and calling the data stored in the memory 502. The memory 502 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created based on the use of the mobile phone, etc. In addition, the memory 502 can include a high-speed random access memory and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0151] Wherein, if the module / unit integrated in the computer device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor 501, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc.

[0152] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the information mining method for easily interviewed users described in any of the above embodiments.

[0153] An embodiment of the present application further provides a computer program product, including a computer program / instruction, which, when executed by a processor of the computer program / instruction, implements the information mining method for easily interviewed users described in any of the above embodiments.

[0154] The methods described herein can be implemented in whole or in part using software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed on a computer, the processes or functions described herein are performed in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user equipment, a core network device, an OAM, or other programmable device.

[0155] The computer program or instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer program or instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless method. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; an optical medium, such as a digital video disk; or a semiconductor medium, such as a solid-state drive. The computer-readable storage medium may be a volatile or non-volatile storage medium, or may include both volatile and non-volatile types of storage media.

[0156] The computer programs / instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in the computer-readable storage medium in each computing / processing device.

[0157] The computer program instructions for performing the operation of the present application can be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data or source code or object code written in any combination of one or more programming languages, wherein the programming language includes object-oriented programming languages such as Smalltalk, C++, and conventional procedural programming languages such as "C" language or similar programming languages. Computer-readable program instructions can be executed completely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or executed completely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer by any type of network including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (such as by using an Internet service provider to connect to the Internet). In certain embodiments, by utilizing the state information of computer-readable program instructions to personalize electronic circuits, such as programmable logic circuits, field programmable gate arrays (FPGAs) or programmable logic arrays (PLAs), the electronic circuits can execute computer-readable program instructions, thereby realizing various aspects of the present application.

[0158] Various aspects of the present application are described herein with reference to flowcharts and / or block diagrams according to embodiments of the present application. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer-readable program instructions.

[0159] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device is generated that implements the functions / actions specified in one or more blocks in the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium, where these instructions cause the computer, programmable data processing device, and / or other device to operate in a specific manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing various aspects of the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0160] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0161] Each box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and the part of the module, program segment or instruction comprises one or more executable instructions for realizing the logical function of regulation. In some alternative implementations, the function annotated in the box can also occur in a sequence different from that annotated in the accompanying drawings. For example, two continuous boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the function involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be realized by a dedicated hardware-based system that performs the function or action of regulation, or can be realized by a combination of dedicated hardware and computer instructions. It is well known to those skilled in the art that realizing by hardware, realizing by software, and realizing by the combination of software and hardware are all equivalent.

[0162] The embodiments of the present application provide a method, apparatus, device, medium and product for mining information of users who are easily interviewed, and the beneficial effects are as follows: by identifying users who are easily interviewed, and constructing a graph structure based on the product themes that are easily interviewed, the product themes of operators, and the frequency of interview recommendations and the frequency of interviews of different product themes, and then based on the information in the graph structure, predicting the frequency of interview recommendations and the frequency of interviews of product themes that are easily interviewed in the future according to the graph convolutional neural network model, it is possible to fully mine the implicit relationship between users, products and operators, improve the prediction accuracy, and thus improve the accuracy of interviewed user information mining; by The product themes of users, the frequency of operators' recommendations for different product themes, and the frequency of product themes with positive emotional trends after the users are interviewed are combined. A graph structure is constructed to mine the frequency characteristics of the three relationships among users, products, and operators that can maintain the positive emotional trends of users. Then, the graph convolutional neural network model is combined to obtain the predicted recommendation frequency and interview frequency. This improves the defect of the existing technology that does not consider the fairness perceived by the interviewed users after the interview, and can enhance the user's product usage experience. In addition, by partitioning the graph structure, the relevance of products can be retained to the maximum extent, and the obtained feature information can be enhanced, thereby improving the prediction accuracy.

[0163] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications are also considered to be within the scope of protection of the present application.

Claims

1. A method for mining information of users who are easily interviewed, characterized in that: include: Identify users who are likely to be interviewed based on user data; Constructing a graph structure based on the product themes that are likely to be of interest to the interviewed users, the product themes of the operators, and the recommendation frequency and interview frequency of different product themes; Based on the information in the graph structure and according to a preset graph convolutional neural network model, the recommendation frequency and the interview frequency of the product topics that are likely to be associated with the interviewed users at a future moment are obtained.

2. The method for mining information of easily interviewed users according to claim 1, wherein: The graph structure is constructed based on the product themes that are likely to be of interest to the interviewed users, the product themes of the operators, and the interview recommendation frequencies and interview frequencies of different product themes, including: For the user who is easily interviewed, obtain the product themes that the user is interested in, the product themes of the operator, the frequency of the operator's interview recommendations for different product themes, and the frequency of the user's positive emotional response to the product themes; Take users, product themes, and operators as nodes, use the interviewed recommendation frequency as the edge between the operator node and the product theme node, use the interviewed frequency as the edge between the user node and the product theme node, and determine the edge between the user node and the operator node based on the interviewed frequency and the interviewed recommendation frequency corresponding to all product themes associated with the user, to obtain a graph structure.

3. The method for mining information of easily interviewed users according to claim 2, wherein: The determining of the edge between the user node and the operator node according to the visit frequency and the visit recommendation frequency corresponding to all product topics associated with the user includes: For each user node, determine whether the frequency of interviews and the frequency of recommendations corresponding to all product topics associated with the user overlap; If yes, the overlapping frequency interval of the interviewed frequency and the interviewed recommendation frequency corresponding to all product themes associated with the user is used as the edge between the user node and the operator node; If not, a frequency interval is determined according to the mean, maximum, and minimum values of the interview frequencies corresponding to all product topics associated with the user, and the frequency interval is used as the edge between the user node and the operator node.

4. The method for mining information of easily interviewed users according to claim 2, wherein: The acquisition of product themes that users are interested in, product themes that operators are interested in, the frequency of operators recommending different product themes, and the frequency of product themes for which users have a positive sentiment after being interviewed include: Extract product topics that users are interested in based on their product reviews; Extract the operator's product theme based on the operator's product description text; Based on the product themes of the operator, determine the number of users corresponding to each product theme according to the similarity between each product theme and the historical product acceptance of each user; obtain the historical frequency of recommendation received by the user, calculate the average of the historical frequency of recommendation received by the number of users corresponding to each product theme, and obtain the frequency of recommendation received by each product theme; Obtain the frequency of users' interviews on different products and the emotional trends of users after each interview, and take the products for which the emotional trends of the users after the interviews are positive as positive emotional products; determine several positive emotional products corresponding to each product theme based on the similarity between the product themes that the users are interested in and the positive emotional products corresponding to the users; calculate the average of the frequency of interviews for the several positive emotional products corresponding to each product theme, and obtain the frequency of interviews for product themes for which the emotional trends of the users after the interviews are positive.

5. The method for mining information of easily interviewed users according to claim 1, wherein: The identifying of susceptible users based on user data includes: Obtaining multi-dimensional user data of the operator, and filtering the multi-dimensional user data to obtain input features; Based on the input features, a user identification model that is susceptible to being interviewed is constructed, and users that are susceptible to being interviewed are obtained according to the user identification model that is susceptible to being interviewed; The characteristic data of the user who is easily interviewed is obtained, and a preset Gaussian mixture model is used to perform classification based on the characteristic data to obtain a category label of the user who is easily interviewed.

6. The method for mining information of easily interviewed users according to claim 1, wherein: The method of obtaining the recommended frequency and interview frequency of the product topics that are likely to be associated with the interviewed users at a future moment based on the information in the graph structure and according to a preset graph convolutional neural network model includes: Partitioning the graph structure according to the distance between each product theme node and an adjacent product theme node in the graph structure to obtain a plurality of subgraph structures; Extracting a feature matrix of product topics in the subgraph structure; The node information in the subgraph structure and the feature matrix are input into a preset graph convolutional neural network model to predict the recommendation frequency and visit frequency of the product topics that are likely to be associated with the interviewed users in the future.

7. An information mining device for users who are easily interviewed, characterized in that: include: A user identification module, used to identify susceptible users based on user data; A graph structure construction module is used to construct a graph structure based on the product themes that are likely to be of interest to the interviewed users, the product themes of the operators, and the frequency of recommendation and interview of different product themes; The model prediction module is used to obtain the recommendation frequency and interview frequency of the product topics that are likely to be associated with the interviewed users at a future moment based on the information in the graph structure and according to a preset graph convolutional neural network model.

8. A computer device, characterized in that: The system comprises a processor and a memory, wherein a computer program is stored in the memory and configured to be executed by the processor, and wherein the method for mining information of susceptible users according to any one of claims 1 to 6 is implemented when the processor executes the computer program.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the device where the computer-readable storage medium is located executes the computer program, the method for mining information of easily interviewed users according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for mining information on susceptible users according to any one of claims 1 to 6 is implemented.

Images