A method and device for predicting intention
By obtaining and analyzing user's personalized and system key feature data in the intelligent customer service system, extracting abnormal and high-frequency mode features, and combining preset rules to make intention predictions, the problem of low accuracy in intention prediction in the existing technology is solved. Improve user experience and prediction accuracy.
Patent Information
- Application Number
- CN202010796464.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-08-10
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2040-08-10
AI Technical Summary
In the prior art, when the user's intention prediction is predicted, the intelligent customer service system has low accuracy in intention prediction, which is difficult to improve the user experience due to the complexity of feature data and the difficulty of maintaining rules.
By obtaining the user's personalized feature data and system key feature data, pattern recognition and statistical analysis are performed, abnormal pattern features and high-frequency mode features are extracted, intention prediction is made in combination with preset rules, and the model is optimized through user feedback.
It significantly improves the accuracy of intention prediction, narrows the range of fitted data, enhances user experience, and further improves prediction accuracy through model optimization and adjustment.
Smart Images

Figure CN113761139B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology, and in particular to a method and device for predicting intention. Background Art
[0002] In the field of intelligent customer service, if the intelligent customer service system can make more accurate predictions about the intentions of new users before they describe their questions, it will significantly improve the user experience. For example, when a user first enters the consultation interface, the intelligent customer service system directly presents the user with a list of possible questions, or directly presents the user with the most likely question-answer pairs, or directly provides the user with the details page of the most likely question (for example, if it is predicted that the user has an 80% probability of inquiring about the order logistics, the logistics progress will be directly presented), etc., all of which can bring a better experience to the user.
[0003] At present, the main technical solutions for intent prediction are to train classification models based on the comprehensive characteristics of the current system to predict intent, or to manually write rules to predict intent based on trigger conditions.
[0004] In the process of implementing the present invention, the inventors found that there are at least the following problems in the prior art:
[0005] 1. For the technical solution of predicting intent through classification models, various comprehensive features in the current system are directly used to train the classification model. Since the various features in the system are too complex and there may be tens of thousands of feature data in the system, it is very difficult to directly predict the user's potential intention from massive data, and the accuracy of model prediction is difficult to improve;
[0006] 2. For the technical solution of predicting intent by manually writing rules, the accuracy of predicting user intent is not high because the rules are difficult to write and maintain, and the applicable scenarios of the rules may become invalid over time. Summary of the invention
[0007] In view of this, an embodiment of the present invention provides a method and device for intention prediction, which can greatly reduce the range of data that needs to be fitted when performing intention prediction, thereby improving the accuracy of intention prediction.
[0008] To achieve the above objective, according to one aspect of an embodiment of the present invention, a method for predicting intention is provided.
[0009] A method for predicting intentions, comprising: obtaining personalized feature data of a user and key feature data of a system; performing pattern recognition on the personalized feature data of the user and the key feature data of the system to obtain abstract features; performing statistical analysis on the abstract features to obtain specified pattern features; and predicting the intentions of the user according to the specified pattern features.
[0010] Optionally, the designated pattern features include abnormal pattern features and high-frequency pattern features, and performing statistical analysis on the abstract features to obtain the designated pattern features includes: performing abnormality detection on the abstract features to obtain abnormal pattern features, and taking features of the abstract features except the abnormal pattern features as normal pattern features; taking features of the normal pattern features whose frequency of occurrence is greater than a set feature frequency threshold as high-frequency pattern features; and taking the abnormal pattern features and the high-frequency pattern features as designated pattern features.
[0011] Optionally, after predicting the user's intention according to the specified pattern characteristics, it also includes: taking the result of predicting the user's intention according to the specified pattern characteristics as a first prediction result; predicting the user's intention according to a preset prediction rule to obtain a second prediction result; and fusing the first prediction result with the second prediction result to predict the user's intention.
[0012] Optionally, fusing the first prediction result with the second prediction result includes: fusing the first prediction result with the second prediction result by setting a priority of the first prediction result and the second prediction result.
[0013] Optionally, it also includes: displaying the result of intention prediction, and collecting feedback data from the user, wherein the feedback data is used to optimize and adjust the intention prediction model.
[0014] Optionally, the training process of the intent prediction model includes: defining abstract feature types related to the business and user intent classification; using an automatic collection mechanism to collect online data in real time to obtain the intent type selected by the user; obtaining the user's personalized feature data and system key feature data, and performing pattern recognition to obtain abstract features; performing statistical analysis on the abstract features to obtain specified pattern features; using the specified pattern features as input and the user-selected intent type as output to train the intent prediction model.
[0015] According to another aspect of an embodiment of the present invention, a device for predicting intention is provided.
[0016] A device for predicting intentions, comprising: a data acquisition module for acquiring personalized feature data of a user and key feature data of a system; a pattern recognition module for performing pattern recognition on the personalized feature data of the user and the key feature data of the system to obtain abstract features; a feature extraction module for performing statistical analysis on the abstract features to obtain specified pattern features; and a feature judgment module for predicting the intentions of the user based on the specified pattern features.
[0017] Optionally, the designated pattern features include abnormal pattern features and high-frequency pattern features, and the feature extraction module is further used to: perform abnormality detection on the abstract features to obtain abnormal pattern features, and use features in the abstract features other than the abnormal pattern features as normal pattern features; use features in the normal pattern features whose frequency of occurrence is greater than a set feature frequency threshold as high-frequency pattern features; and use the abnormal pattern features and the high-frequency pattern features as designated pattern features.
[0018] Optionally, it also includes a second prediction module, which is used to: after predicting the user's intention according to the specified pattern characteristics, use the result of predicting the user's intention according to the specified pattern characteristics as the first prediction result; predict the user's intention according to a preset prediction rule to obtain a second prediction result; and merge the first prediction result with the second prediction result to predict the user's intention.
[0019] Optionally, the second prediction module is further used to: merge the first prediction result and the second prediction result by setting the priority of the first prediction result and the second prediction result.
[0020] Optionally, it also includes a prediction interaction module, which is used to: display the results of intention prediction and collect feedback data from the user, and the feedback data is used to optimize and adjust the intention prediction model.
[0021] Optionally, the training process of the intent prediction model includes: defining abstract feature types related to the business and user intent classification; using an automatic collection mechanism to collect online data in real time to obtain the intent type selected by the user; obtaining the user's personalized feature data and system key feature data, and performing pattern recognition to obtain abstract features; performing statistical analysis on the abstract features to obtain specified pattern features; using the specified pattern features as input and the user-selected intent type as output to train the intent prediction model.
[0022] According to yet another aspect of an embodiment of the present invention, an electronic device for predicting intention is provided.
[0023] An electronic device for intention prediction includes: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method for intention prediction provided by an embodiment of the present invention.
[0024] According to still another aspect of the embodiments of the present invention, a computer readable medium is provided.
[0025] A computer-readable medium stores a computer program, which, when executed by a processor, implements the method for predicting intention provided by an embodiment of the present invention.
[0026] An embodiment of the above invention has the following advantages or beneficial effects: by obtaining the personalized feature data of the user and the key feature data of the system; performing pattern recognition on the personalized feature data of the user and the key feature data of the system to obtain abstract features; performing statistical analysis on the abstract features to obtain specified pattern features; and predicting the user's intention according to the specified pattern features, it is possible to greatly reduce the data range that needs to be fitted when predicting the intention, thereby improving the accuracy of the intention prediction. According to the technical solution of the embodiment of the present invention, it is possible to realize the discovery of abnormal pattern features and high-frequency pattern features that appear during the operation of the system based on the abstract features obtained by pattern recognition, and these features can better reflect the needs of the user, so that the user intention prediction based on these features can improve the accuracy of the intention prediction; and, it is also possible to perform intention prediction in combination with the set prediction rules, thereby further improving the accuracy of the intention prediction; in addition, after the intention prediction results are obtained, the user's feedback data is collected, and the model is optimized and adjusted through the user feedback data, thereby further improving the accuracy of the intention prediction model.
[0027] The further effects of the above-mentioned non-conventional optional manner will be described below in conjunction with the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.
[0029] Figure 1 is a schematic diagram of main steps of a method for predicting intention according to an embodiment of the present invention;
[0030] Figure 2 Schematic diagram of the implementation principle of the intention prediction system of an embodiment of the present invention;
[0031] Figure 3 is a schematic diagram of main modules of a device for intention prediction according to an embodiment of the present invention;
[0032] Figure 4 is an exemplary system architecture diagram to which embodiments of the present invention may be applied;
[0033] Figure 5 It is a schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing an embodiment of the present invention. DETAILED DESCRIPTION
[0034] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.
[0035] At present, the main technical solutions for predicting user intentions are as follows:
[0036] 1. Train a classification model based on the comprehensive features of the current system to predict intent
[0037] At present, the typical approach is to treat intent prediction as a classification problem. Intent prediction is regarded as the mapping relationship between hundreds of system features and question classification. The general process is as follows: from historical conversation messages, information such as user consultation questions and question labels are mined, and various user characteristics are traced from the system, such as user gender, age range, frequently purchased products, order status, logistics status, consultation status, etc. data are used to fit the relationship between the original features and question labels through training models;
[0038] 2. Manually write rules to predict intent based on trigger conditions
[0039] If there are obvious rules that can hit the user's intention with a high probability, you can also configure rules to predict the user's intention. For example, if a user's order has not been delivered for more than 3 days, and the user clicks on the inquiry page to inquire about the logistics of the order, it is likely that the user is inquiring about the logistics of the order.
[0040] However, the above two solutions have the defect of low prediction accuracy. In order to solve the technical problem existing in the prior art, the present invention provides a method and device for intention prediction, which mainly improves the prediction accuracy of user intention from the perspective of pattern recognition of personalized data and overall system basic data. The main idea is: no longer regard the prediction process as a direct mapping relationship between system features and intentions, but to discover abnormal patterns or high-frequency patterns from the user's personalized data and the overall operation status of the system, and then deduce the user's potential intentions from the abnormal patterns or high-frequency patterns. This can greatly reduce the data range that the model needs to fit, thereby improving the accuracy of intention prediction.
[0041] The entire system can be constructed as an automated execution solution. First, it can monitor the standard and high-frequency patterns of the system's overall data in real time, analyze the typical problem types in the manual consultation data, count the high-frequency patterns and abnormal patterns, and automatically count the correspondence between the pattern types and the consultation problem types. This can achieve the purpose of automatic dynamic adjustment and avoid being fixed like manual rules.
[0042] In the introduction to the embodiments of the present invention, the terms that appear are explained as follows:
[0043] Intent: Intent, when consulting with intelligent customer service, indicates the purpose of the user's consultation;
[0044] Prediction: predict, before the user makes a consultation, pre-judge the user's consultation intention;
[0045] Pattern: A pattern is a regularity that exists in the real world or in human designs or abstract ideas;
[0046] Features: feature, as input parameters for model training, such as order volume, duration, etc.;
[0047] Model: Model is a mathematical model that describes the relationship between features and outputs. It can output corresponding probability values based on a set of input features.
[0048] Anomaly detection: First, define a set of data when the system is in a "normal" state, such as CPU utilization, memory utilization, file checksums, etc., and then analyze it to determine whether anomalies occur;
[0049] XGBoost: is an optimized distributed gradient boosting model library that is very efficient, flexible and portable. It is very simple and powerful in the field of structured feature training.
[0050] CNN: convolutional neural network, convolutional neural network.
[0051] Figure 1 Schematic diagram of the main steps of the method for intention prediction according to an embodiment of the present invention. Figure 1 As shown, the method for intention prediction according to the embodiment of the present invention mainly includes the following steps S101 to S104.
[0052] Step S101: Acquire user's personalized feature data and system key feature data;
[0053] Step S102: performing pattern recognition on the user's personalized feature data and the system's key feature data to obtain abstract features;
[0054] Step S103: performing statistical analysis on the abstract features to obtain designated pattern features;
[0055] Step S104: Predict the user's intention based on the specified pattern features.
[0056] According to the above-mentioned steps S101 to S104, the specified pattern features can be discovered by processing the user's personalized feature data and the system's key feature data, and intention prediction can be performed based on the specified pattern features. This can greatly reduce the data range that needs to be fitted when predicting intention, thereby improving the accuracy of intention prediction.
[0057] According to an embodiment of the present invention, the specified pattern features include abnormal pattern features and high-frequency pattern features, and performing statistical analysis on the abstract features to obtain the specified pattern features may specifically include the following steps:
[0058] Perform anomaly detection on the abstract features to obtain abnormal pattern features, and use the features in the abstract features except the abnormal pattern features as normal pattern features;
[0059] Among the normal mode features, the features whose frequency of occurrence is greater than the set feature frequency threshold are regarded as high-frequency mode features;
[0060] Abnormal pattern features and high-frequency pattern features are taken as designated pattern features.
[0061] According to the technical solution of this embodiment, it is possible to realize the abstract features obtained based on pattern recognition and discover abnormal pattern features and high-frequency pattern features that appear during the operation of the system. These features can better reflect the needs of users, and thus predicting user intentions based on these features can improve the accuracy of intention prediction.
[0062] According to another embodiment of the present invention, after predicting the user's intention according to the specified pattern feature, the method may further include:
[0063] The result of predicting the user's intention according to the specified pattern feature is used as the first prediction result;
[0064] Predicting the user's intention according to a preset prediction rule to obtain a second prediction result;
[0065] The first prediction result is combined with the second prediction result to predict the user's intention.
[0066] According to this embodiment, the first prediction result of the intention prediction through the intention prediction algorithm model can be integrated with the second prediction result of the intention prediction according to the set prediction rule, so as to further improve the accuracy of the intention prediction. Among them, the preset prediction rule is set according to the business situation. For example, if the user's order has not been delivered for more than 3 days, and the user clicks to enter the consultation from the order page, the user's intention is predicted to consult the logistics situation of the order.
[0067] According to an embodiment of the present invention, when the first prediction result and the second prediction result are merged, one implementation method is, for example: by setting the priority of the first prediction result and the second prediction result, the first prediction result and the second prediction result are merged.
[0068] According to another embodiment of the present invention, after the intention prediction result is obtained, the intention prediction result can also be displayed, and the user's feedback data can be collected, and the feedback data is used to optimize and adjust the intention prediction model. After the intention prediction result is displayed, if the user clicks on the result, or the user actively feedbacks that the problem is "solved" or "unsolved", these can be collected as user feedback data, and the intention prediction model can be optimized and adjusted based on these feedback data. In this way, the model can be adjusted through user feedback data, thereby improving the accuracy of the intention prediction model.
[0069] In an embodiment of the present invention, the training process of the intention prediction model may include the following steps:
[0070] Define abstract feature types related to the business and user intent classification;
[0071] Use automatic collection mechanisms to collect online data in real time to obtain the user's selected intent type;
[0072] Obtain the user's personalized feature data and system key feature data, and perform pattern recognition to obtain abstract features;
[0073] Statistical analysis of the abstract features to obtain the specified pattern characteristics;
[0074] The specified pattern features are used as input and the user-selected intent type is used as output to train the intent prediction model.
[0075] The implementation process of the intention prediction system of the present invention is described below in conjunction with specific embodiments. Figure 2 Schematic diagram of the implementation principle of the intention prediction system of the embodiment of the present invention. Figure 2 As shown, the intention prediction system of the embodiment of the present invention mainly includes a user data management module, a system key feature data management module, a pattern recognition and feature extraction module, an intention prediction module, a rule maintenance module and a prediction interaction module. Among them:
[0076] User data management module: used to define a set of observation features that are strongly related to user consultation intentions, such as user age, gender, frequently purchased list, consumption level, credit, comment features, customer service consultation frequency, etc.
[0077] System key feature data management module: used to define a set of key data features that affect the overall system, reflecting the system status as a whole, such as the latest promotional information of the e-commerce system, large-scale coupon anomalies, and other status information;
[0078] Pattern recognition and feature extraction module: mainly responsible for statistical system key feature data and user personalized feature data, such as the overall order volume of the system, logistics distribution volume, average user order volume, user consultation volume, user complaint volume, the number and ratio of single user order and consultation time, single user consultation frequency, and consultation frequency of a product on a single day. Then, pattern recognition is performed on these feature data to obtain the data range distribution characteristics of each abstract feature. After that, the abnormal data of each abstract feature is obtained through anomaly detection methods (such as IQR, Z-score, etc.), and these abnormal data are used as abnormal pattern features, and the remaining feature data within the normal fluctuation range of each abstract feature is used as normal pattern features; then, according to the frequency of occurrence of each abstract feature, high-frequency pattern features are obtained from the normal pattern features, and the abnormal pattern features and high-frequency pattern features are used as designated pattern features for further processing. For example, if the number of orders for a certain mobile phone between 3pm and 5pm is between 1000 and 1200, then for the abstract feature of the number of orders for a certain mobile phone between 3pm and 5pm, its normal mode feature value is between 1000 and 1200. High-frequency mode belongs to the data within the normal mode range, but has stronger statistical characteristics. For example, the number of orders for a certain mobile phone between 3pm and 5pm is between 1000 and 1200, and 200 of them are sold with mobile phone cases. Among them, IQR (interquartile range) is the interquartile range, also known as the four-point difference. It is a method in descriptive statistics to determine the difference between the third quartile and the first quartile. Like the variance and standard deviation, it indicates the dispersion of each variable in the statistical data; Z-score is the Z score, also called the standard score, which is the process of dividing the difference between a number and the mean by the standard deviation. In statistics, the standard score is the symbol number of the standard deviation of the value of an observation or data point above the mean of the observed value or measurement;
[0079] Intention prediction module: used to judge the user's high-frequency intentions through classification models based on the processing results of the pattern recognition and feature extraction modules;
[0080] Rule maintenance module: used to maintain preset intention prediction rules and determine whether the user's customer service consultation behavior complies with the intention prediction rules. Intention prediction rules are used as a supplementary means in cases where strong types or direct control are required;
[0081] Prediction interaction module: responsible for interacting with front-end users, including reading the prediction results of the intent prediction module and the prediction results of the rule maintenance module when the user first enters the customer service consultation, and showing the user the similar questions corresponding to the predicted intent, or the high-probability consultation question and answer pairs, or the further details of the high-probability intent. If the user clicks the question or feedback button, the user's click data is collected as user feedback, and the user feedback is fed back to the intent prediction module to further improve the accuracy of the prediction model. In addition, statistical analysis can be performed on user feedback to analyze obvious policy problems, such as a rule that has been set unreasonably, and then the rule can be adjusted to better improve the accuracy of intent prediction.
[0082] The implementation principle and training process of the intent prediction model of the embodiment of the present invention will be introduced below. The intent prediction model of the embodiment of the present invention is essentially a classification model, but unlike the traditional direct classification model, it does not directly use the original data features to fit the user's intent, but uses the recognition results of the pattern recognition and feature extraction module to make judgments. The pattern recognition and feature extraction module first determines the pattern range and outputs whether it is abnormal for the main features (including system key features and user personalized features). In this way, abnormal pattern features and high-frequency pattern features are obtained, and these features are used to fit the common intent classification labels of user consultations.
[0083] Assume that the original features are features such as the number of user orders and the waiting time for users to place orders, and after being processed by the pattern recognition and feature extraction modules, they are converted into higher-level and clearer features. In the embodiment of the present invention, they are designated pattern features such as abnormal pattern features and high-frequency pattern features, for example:
[0084] 1) Whether the order delivery time exceeds the normal range (e.g. 90% of users receive successful delivery within 24 hours);
[0085] 2) Whether the order address is different from the user's general address;
[0086] 3) Whether the user has triggered the order reminder operation, the number and frequency of the order reminder operation, and whether there are any abnormalities in the processing cycle;
[0087] 4) The user’s most recent consultation topic classification, whether it triggers a complaint, etc.
[0088] When training the intent prediction model of the present invention, it is first necessary to perform feature definition and intent classification. In combination with the business situation, abstract feature types in hundreds of fields closely related to the business are pre-planned, and features that can obtain results from pattern recognition that is closely related to customer service consultation are selected. Such as "whether the order delivery time exceeds the normal range" in the above embodiment. Among all possible intent classifications of user consultations, screen out intent classifications that can be predicted in advance, such as logistics status inquiry, refund status consultation, application for price protection operations, etc.
[0089] Then, collect training sample data. You can first manually annotate a batch of basic data, and use an automatic collection mechanism to collect data. Set up a real-time collection mechanism. For example, when a user triggers the selected intent type during consultation, find the user's current various basic data, and perform feature abstraction processing and feature statistical analysis through the pattern recognition and feature extraction module to obtain the specified pattern features. After a period of time, when a sufficient amount of specified pattern features and intent classification labels (such as millions or tens of millions of sets of training data) are collected, they can be used as training samples to train the intent prediction model.
[0090] Since the specified pattern features and intent classification labels in the training sample data are mainly structured data, the XGBoost model can be used to fit the classification relationship, or a deep learning model (such as CNN, etc.) can be used to implement the model structure. The specified pattern features are used as input, and the intent type selected by the user is used as output to train the intent prediction model.
[0091] The following describes the implementation process of online intention prediction by the intention prediction system of the embodiment of the present invention. In the embodiment of the present invention, taking the customer service consultation system of the e-commerce platform to predict the user's intention as an example, the online prediction process needs to be combined with Figure 2 The modules in the system are connected with the overall system data and the user's personalized data, which mainly includes the following steps:
[0092] 1. When a user enters customer service consultation, the intention prediction system is triggered and executed. The prediction interaction module reads the user's recent personalized data, such as order quantity, purchased product category, product name, logistics, consultation history, etc., as well as key feature data of the system, such as the current promotion information of the system, the coupon information of the system, etc.;
[0093] 2. Obtain the specified pattern features of the current user and system through the pattern recognition and feature extraction module, including abnormal pattern features and high-frequency pattern features;
[0094] 3. Input the specified pattern features of the current user and the system into the intention prediction model, and output the classification result of the intention prediction model as the first prediction result;
[0095] 4. Obtain the intention prediction result of the rule maintenance module as the second prediction result;
[0096] 5. Fusing the first prediction result and the second prediction result. The results can be fused according to a certain strategy. For example, the results can be fused by setting the priority of the first prediction result and the second prediction result. For example, the results can be fused by selecting a threshold + rule priority method.
[0097] 6. The prediction interaction module displays the questions, question-answer pairs, or content details of the N most likely intent prediction results to the user (for specific content, different strategies can be configured according to different intent classifications);
[0098] 7. Use the intention prediction results of whether the user clicks or views the push as feedback data, collect the feedback data and send it to the intent prediction model, so as to further improve the accuracy of the model and strategy.
[0099] Among them, in the embodiment of the present invention, when the first prediction result and the second prediction result are merged in the above step 5, if the threshold + rule priority method is selected to merge the results, its possible implementation methods include, for example:
[0100] 1) Assign high, medium, and low priorities to the preset intent prediction rules. If a high-priority rule is triggered, the prediction result corresponding to the high-priority rule is displayed unconditionally at the front position;
[0101] 2) Specify a threshold for the predicted intention score output by the intention prediction model, take the predicted intention above the threshold as the prediction result, and arrange it after the prediction result corresponding to the high priority rule for display;
[0102] 3) If there is no hit intent prediction rule, but only the predicted intent output by the intent prediction model, then the multiple predicted intents output by the intent prediction model are sorted according to the predicted intent scores and displayed as the prediction results;
[0103] 4) If the intent prediction rules and the outputs of the intent prediction model are mixed, the results are displayed based on the priority of the set rules and the priority of the predicted intent results of the intent prediction model.
[0104] Figure 3 Schematic diagram of the main modules of the device for intention prediction according to an embodiment of the present invention. Figure 3 As shown, the device 300 for intention prediction according to the embodiment of the present invention mainly includes a data acquisition module 301 , a pattern recognition module 302 , a feature extraction module 303 and a feature judgment module 304 .
[0105] Data acquisition module 301, used to acquire user's personalized feature data and system key feature data;
[0106] A pattern recognition module 302 is used to perform pattern recognition on the user's personalized feature data and the system key feature data to obtain abstract features;
[0107] A feature extraction module 303 is used to perform statistical analysis on the abstract features to obtain specified pattern features;
[0108] The feature judgment module 304 is used to predict the user's intention based on the specified pattern features.
[0109] According to an embodiment of the present invention, the specified pattern features include abnormal pattern features and high-frequency pattern features, and the feature extraction module 303 can also be used to:
[0110] Performing anomaly detection on the abstract features to obtain abnormal pattern features, and taking features of the abstract features other than the abnormal pattern features as normal pattern features;
[0111] Among the normal mode features, features whose frequency of occurrence is greater than a set feature frequency threshold are taken as high-frequency mode features;
[0112] The abnormal pattern features and the high-frequency pattern features are used as designated pattern features.
[0113] According to another embodiment of the present invention, the device 300 for predicting intention may further include a second prediction module (not shown in the figure) for:
[0114] After predicting the user's intention according to the specified pattern feature, using the result of predicting the user's intention according to the specified pattern feature as a first prediction result;
[0115] Predicting the user's intention according to a preset prediction rule to obtain a second prediction result;
[0116] The first prediction result is combined with the second prediction result to predict the intention of the user.
[0117] According to another embodiment of the present invention, the second prediction module (not shown in the figure) can also be used for:
[0118] The first prediction result and the second prediction result are merged by setting the priority of the first prediction result and the second prediction result.
[0119] According to another embodiment of the present invention, the intention prediction device 300 may further include a prediction interaction module (not shown in the figure) for:
[0120] The result of intention prediction is displayed, and feedback data from the user is collected, where the feedback data is used to optimize and adjust the intention prediction model.
[0121] According to yet another embodiment of the present invention, the training process of the intention prediction model may include:
[0122] Define abstract feature types related to the business and user intent classification;
[0123] Use automatic collection mechanisms to collect online data in real time to obtain the user's selected intent type;
[0124] Obtain the user's personalized feature data and system key feature data, and perform pattern recognition to obtain abstract features;
[0125] Performing statistical analysis on the abstract features to obtain specified pattern features;
[0126] The designated pattern feature is used as input and the user-selected intent type is used as output to train the intent prediction model.
[0127] According to the technical solution of the embodiment of the present invention, by obtaining the personalized feature data of the user and the key feature data of the system; performing pattern recognition on the personalized feature data of the user and the key feature data of the system to obtain abstract features; performing statistical analysis on the abstract features to obtain specified pattern features; and predicting the user's intention according to the specified pattern features, it is possible to greatly reduce the data range that needs to be fitted when performing intention prediction, thereby improving the accuracy of intention prediction. According to the technical solution of the embodiment of the present invention, it is possible to realize the abstract features obtained based on pattern recognition, discover abnormal pattern features and high-frequency pattern features that appear during the operation of the system, and these features can better reflect the needs of the user, so that the user intention prediction based on these features can improve the accuracy of intention prediction; and, it is also possible to perform intention prediction in combination with the set prediction rules, thereby further improving the accuracy of intention prediction; in addition, after the intention prediction results are obtained, the user's feedback data is collected, and the model is optimized and adjusted through the user feedback data, thereby further improving the accuracy of the intention prediction model.
[0128] Figure 4 An exemplary system architecture 400 is shown to which the method or apparatus for prejudging intention according to an embodiment of the present invention may be applied.
[0129] like Figure 4As shown, system architecture 400 may include terminal devices 401, 402, 403, network 404 and server 405. Network 404 is used to provide a medium for communication links between terminal devices 401, 402, 403 and server 405. Network 404 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.
[0130] Users can use terminal devices 401, 402, 403 to interact with server 405 through network 404 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 401, 402, 403, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only as examples).
[0131] The terminal devices 401 , 402 , and 403 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.
[0132] The server 405 may be a server that provides various services, such as a backend management server (only an example) that provides support for shopping websites browsed by users using the terminal devices 401, 402, and 403. The backend management server may analyze and process the received data such as product information query requests, and feed back the processing results (such as target push information, product information - only an example) to the terminal device.
[0133] It should be noted that the intention prediction method provided in the embodiment of the present invention is generally executed by the server 405 , and accordingly, the intention prediction device is generally set in the server 405 .
[0134] It should be understood that Figure 4 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.
[0135] Reference below Figure 5 , which shows a schematic diagram of the structure of a computer system 500 of a terminal device or a server suitable for implementing an embodiment of the present invention. Figure 5 The terminal device or server shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.
[0136] like Figure 5As shown, the computer system 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage part 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the system 500 are also stored. The CPU 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.
[0137] The following components are connected to the I / O interface 505: an input section 506 including a keyboard, a mouse, etc.; an output section 507 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, a modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the I / O interface 505 as needed. A removable medium 511, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 510 as needed, so that a computer program read therefrom is installed into the storage section 508 as needed.
[0138] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 509, and / or installed from the removable medium 511. When the computer program is executed by the central processing unit (CPU) 501, the above-mentioned functions defined in the system of the present invention are executed.
[0139] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.
[0140] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0141] The units or modules involved in the embodiments of the present invention may be implemented by software or hardware. The units or modules described may also be provided in a processor. For example, they may be described as follows: a processor includes a data acquisition module, a pattern recognition module, a feature extraction module, and a feature judgment module. The names of these units or modules do not, in certain cases, constitute limitations on the units or modules themselves. For example, the data acquisition module may also be described as a "module for acquiring personalized feature data of users and key feature data of the system."
[0142] As another aspect, the present invention further provides a computer-readable medium, which may be included in the device described in the above embodiment; or may exist independently without being assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by a device, the device includes: obtaining the personalized feature data of the user and the key feature data of the system; performing pattern recognition on the personalized feature data of the user and the key feature data of the system to obtain abstract features; performing statistical analysis on the abstract features to obtain specified pattern features; and predicting the intention of the user according to the specified pattern features.
[0143] According to the technical solution of the embodiment of the present invention, by obtaining the personalized feature data of the user and the key feature data of the system; performing pattern recognition on the personalized feature data of the user and the key feature data of the system to obtain abstract features; performing statistical analysis on the abstract features to obtain specified pattern features; and predicting the user's intention according to the specified pattern features, it is possible to greatly reduce the data range that needs to be fitted when performing intention prediction, thereby improving the accuracy of intention prediction. According to the technical solution of the embodiment of the present invention, it is possible to realize the abstract features obtained based on pattern recognition, discover abnormal pattern features and high-frequency pattern features that appear during the operation of the system, and these features can better reflect the needs of the user, so that the user intention prediction based on these features can improve the accuracy of intention prediction; and, it is also possible to perform intention prediction in combination with the set prediction rules, thereby further improving the accuracy of intention prediction; in addition, after the intention prediction results are obtained, the user's feedback data is collected, and the model is optimized and adjusted through the user feedback data, thereby further improving the accuracy of the intention prediction model.
[0144] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for predicting intention, characterized in that: include: Obtaining the user's personalized feature data and the system's key feature data, wherein the user's personalized feature data is an observation feature that is strongly related to the user's consultation intention, and the system's key feature data is feature data that reflects the system status as a whole; Performing pattern recognition on the personalized feature data of the user and the key feature data of the system to obtain abstract features, wherein the abstract features are selected based on predefined abstract feature types related to the business; Performing statistical analysis on the abstract features to obtain specified pattern features; Predicting the user's intention according to the specified pattern features; The designated pattern features include abnormal pattern features and high-frequency pattern features, and performing statistical analysis on the abstract features to obtain the designated pattern features includes: performing abnormality detection on the abstract features to obtain abnormal pattern features, and taking features of the abstract features other than the abnormal pattern features as normal pattern features; Among the normal mode features, features whose frequency of occurrence is greater than a set feature frequency threshold are taken as high-frequency mode features; and the abnormal mode features and the high-frequency mode features are taken as designated mode features.
2. The method according to claim 1, characterized in that After predicting the user's intention according to the specified mode feature, the method further includes: using a result of predicting the user's intention according to the specified mode feature as a first prediction result; Predicting the user's intention according to a preset prediction rule to obtain a second prediction result; The first prediction result is combined with the second prediction result to predict the intention of the user.
3. The method according to claim 2, characterized in that The step of fusing the first prediction result with the second prediction result includes: The first prediction result and the second prediction result are merged by setting the priority of the first prediction result and the second prediction result.
4. The method according to claim 1 or 2, characterized in that: Also includes: The result of intention prediction is displayed, and feedback data from the user is collected, where the feedback data is used to optimize and adjust the intention prediction model.
5. The method according to claim 4, characterized in that The training process of the intention prediction model includes: Define abstract feature types related to the business and user intent classification; Use automatic collection mechanisms to collect online data in real time to obtain the user's selected intent type; Obtain the user's personalized feature data and system key feature data, and perform pattern recognition to obtain abstract features; Performing statistical analysis on the abstract features to obtain specified pattern features; The designated pattern feature is used as input and the user-selected intent type is used as output to train the intent prediction model.
6. A device for predicting intention, characterized in that: include: A data acquisition module, used to acquire the user's personalized feature data and system key feature data, wherein the user's personalized feature data is an observation feature that is strongly related to the user's consultation intention, and the system key feature data is feature data that reflects the system status as a whole; A pattern recognition module, used for performing pattern recognition on the personalized feature data of the user and the key feature data of the system to obtain abstract features, wherein the abstract features are features selected based on predefined abstract feature types related to the business; A feature extraction module, used for performing statistical analysis on the abstract features to obtain specified pattern features; A feature judgment module, used to predict the user's intention based on the specified pattern features; The designated pattern features include abnormal pattern features and high-frequency pattern features, and the feature extraction module is also used to: perform abnormality detection on the abstract features to obtain abnormal pattern features, and use the features of the abstract features other than the abnormal pattern features as normal pattern features; use the features of the normal pattern features whose frequency of occurrence is greater than a set feature frequency threshold as high-frequency pattern features; and use the abnormal pattern features and the high-frequency pattern features as designated pattern features.
7. The device according to claim 6, characterized in that It also includes a second prediction module, which is used for: After predicting the user's intention according to the specified pattern feature, using the result of predicting the user's intention according to the specified pattern feature as a first prediction result; Predicting the user's intention according to a preset prediction rule to obtain a second prediction result; The first prediction result is combined with the second prediction result to predict the intention of the user.
8. The device according to claim 7, characterized in that The second prediction module is also used for: The first prediction result and the second prediction result are merged by setting the priority of the first prediction result and the second prediction result.
9. The device according to claim 6 or 7, characterized in that It also includes a prejudgment interaction module for: The result of intention prediction is displayed, and feedback data from the user is collected, where the feedback data is used to optimize and adjust the intention prediction model.
10. The device according to claim 9, characterized in that The training process of the intention prediction model includes: Define abstract feature types related to the business and user intent classification; Use automatic collection mechanisms to collect online data in real time to obtain the user's selected intent type; Obtain the user's personalized feature data and system key feature data, and perform pattern recognition to obtain abstract features; Performing statistical analysis on the abstract features to obtain specified pattern features; The designated pattern feature is used as input and the user-selected intent type is used as output to train the intent prediction model.
11. An electronic device for predicting intention, characterized in that: include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 5.
12. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.
