Feature analysis method, device and medium based on continuous interaction

Abstract

The present invention relates to the field of communication technology, and is particularly suitable for the analysis of multiple interactions in intelligent communication interactions. In view of the current lack of analysis of scenarios for multiple calls, the present invention proposes a feature analysis method, device, and medium based on continuous interaction, aiming to solve the technical problem of how to perform change analysis on the information of single-call and / or multi-call interaction records continuously generated in communication interactions, so as to simply and efficiently implement feature analysis, facilitate prediction of interaction trends, intelligent communication control, evaluation of interaction status, etc. To this end, the method of the present invention performs grouping and feature calculation according to predetermined conditions based on the continuously generated interaction information, and determines feature data through spliced ​​features, so as to judge trends, control communications, evaluate interaction status, etc., which is conducive to efficient, convenient, and accurate subsequent prediction control and evaluation, and improves user experience.

Description

Technical Field

[0001] The present invention belongs to the field of communication technology, and is particularly suitable for multiple interaction analysis in communication interactions. More specifically, it relates to a feature analysis method, device and medium based on continuous interaction. Background Art

[0002] With the development of communication technology and artificial intelligence technology, people, machines, and machines can communicate with each other through various intelligent or non-intelligent terminals, such as audio and video calls, text calls, or communication interactions, two-way communication, etc. For example, telephone sales, referred to as telemarketing, can use the terminal to communicate with customers in the name of a company through auxiliary methods such as the Internet, fax, SMS, and mail delivery, and use dedicated marketing numbers (such as telemarketing numbers) and other methods, and use automated information management technology and professional operation platforms to complete product recommendations, consultations, quotations, and confirmation of product transaction conditions; in this way, when telemarketing is needed, customer service and target objects (potential or actual customers) can communicate well. Through the analysis of the characteristics of multiple calls or communication interaction records / information that last for a period of time, it is possible to judge / predict the trend / development of interactive communication, evaluate the status of interactive communication, and control whether communication resources need to be occupied to continue interactive communication in the future. In the prior art, however, feature analysis is usually performed for a single call. For scenarios where multiple calls are made, such as calling a potential customer three times in a week to recommend courses and other communication interactions, only single calls can be analyzed. The dynamic and continuously increasing feature data and its dimensions for each call will change accordingly. It is not possible to easily and efficiently complete feature analysis to determine feature data of fixed dimensions, and it is not possible to make effective use of the changes in communication interactions over a period of time to assist in subsequent prediction trends, evaluation status, and control of communication resource usage. For example, for a customer service robot in a telemarketing scenario, subsequent communication follow-up execution interactions are always performed with a single call. It is not possible to make accurate and efficient judgments on whether to continue to follow up and execute interactions in situations where multiple calls are actually made by the same potential customer. Communication resources are continued to be occupied for situations that do not require follow-up. At the same time, it also lacks defects in the communication interaction method, such as accurate responses, efficient communication, and determination of customer intentions. Obviously, the feature analysis and feature data generation of multiple interactions in the prior art are inefficient, complex, and have poor user experience.

[0003] Therefore, it is necessary to improve the existing analysis method and provide a solution that can achieve efficient and accurate feature analysis and determine fixed-dimensional feature data for multiple communication interactions over a period of time, so as to improve the effective control and control accuracy of intelligent communication control, better predict the development and change trend of communication interaction, and more accurately and efficiently evaluate the interaction status. Especially in the case of intelligent human-machine communication interaction, in the sustainable and dynamically increasing call environment, accurate and efficient control and management of communication interaction can be achieved, communication resource usage can be reduced, and user experience can be improved. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] The present invention aims to solve the technical problem of how to perform change analysis on the information of single-channel and / or multi-channel communication interaction records generated over a period of time; further, to solve the technical problem of how to generate / determine feature data (dimension) based on continuous interaction by means of feature analysis of grouped calculations; further, to solve the technical problem of how to judge trends by separately calculating the features of the groups and generating feature data based on the feature splicing.

[0006] (II) Technical solution

[0007] To solve the above technical problems, the first aspect of the present invention proposes a feature analysis method based on continuous interaction, including: obtaining target interaction content; grouping interaction content belonging to the same target in the interaction content; respectively calculating the features of the interactions in each group of interaction content; and splicing the features of each group to generate feature data.

[0008] According to a preferred embodiment of the present invention, obtaining target interaction content specifically includes: selecting any continuous time period as a continuous period of time for analysis according to analysis requirements, obtaining all interactions within the continuous period of time, and determining multiple interactions belonging to the same target among all interactions; grouping the interaction content belonging to the same target in the interaction content specifically includes: dividing the multiple interactions belonging to the same target into two groups according to predetermined conditions, wherein the multiple interactions include each interaction information recorded corresponding to the multiple interactions.

[0009] According to a preferred embodiment of the present invention, it also includes: accumulating the number of the multiple interactions belonging to the same target in chronological order within the continuous period of time, and extracting the multiple interactions; the predetermined conditions include: the accumulated number of times is an odd number or an even number; the interaction information includes the time of the interaction.

[0010] According to a preferred embodiment of the present invention, the multiple interactions belonging to the same target are divided into two groups according to predetermined conditions, specifically including: when the accumulated number of times is an odd number 2n+1, the first n interactions and the n+1th interaction are the first group, and the n+1th interaction to the 2n+1th interaction are the second group; when the accumulated number of times is an even number 2n, the first n interactions are the first group, and the n+1th interaction to the 2nth interaction are the second group; wherein n is an integer greater than or equal to 0.

[0011] According to a preferred embodiment of the present invention, the features of the interactions in each group of interactive content are calculated separately, specifically including: obtaining features from the text information of each interaction in each group; adding the features of all interactions in each group and then taking the average value; extracting features of a preset fixed dimension from the average value; splicing the features of each group to generate feature data, specifically including: adding the features of the fixed dimension extracted from each group to generate feature data of twice the fixed dimension.

[0012] According to a preferred embodiment of the present invention, it also includes: the text information is text information input during interaction, audio is converted into text information through FR or voice recognition during interaction, and / or video image is converted into text information through character recognition or image recognition during interaction; the feature is a multidimensional feature vector; the feature vector is obtained from the text information based on word segmentation, keywords, and / or word frequency calculation.

[0013] According to a preferred embodiment of the present invention, it also includes: the interaction content is multiple calls; the voice of the call is converted into text to obtain text information corresponding to the call; the target is the customer in the interaction; obtaining the target interaction content also includes: obtaining multiple calls with the customer in all logs over a period of time, and determining the multiple calls belonging to the same customer based on the customer identity identifier.

[0014] According to a preferred embodiment of the present invention, it also includes: constructing a prediction model for judging the interaction development trend, and inputting the characteristic data into the trained prediction model; judging whether the interaction development trend is positive or negative according to the output prediction result, so as to determine the interaction development trend; and / or, judging the interaction development trend according to the characteristic data, and controlling whether it is necessary to continue to occupy communication resources and interact with the target communication; and / or, constructing an evaluation model for evaluating the quality of the interaction state, inputting the characteristic data into the trained evaluation model, and judging whether the interaction is in different scoring states according to the evaluation result.

[0015] To solve the above technical problems, the second aspect of the present invention proposes a feature analysis device based on continuous interaction, including: a grouping module, used to obtain target interaction content and group the interaction content belonging to the same target in the interaction content; a generation module, used to respectively calculate the features of the interactions in each group of interaction content, splice the features of each group, and generate feature data.

[0016] To solve the above technical problems, the third aspect of the present invention proposes an electronic device, comprising a processor and a memory, wherein the memory is used to store a computer executable program, and when the computer program is executed by the processor, the processor executes the method described in any one of the first aspects.

[0017] In order to solve the above technical problems, the fourth aspect of the present invention proposes a computer-readable medium storing a computer executable program, and when the computer executable program is executed, it implements the method described in any one of the first aspects.

[0018] In order to solve the above technical problem, the fifth aspect of the present invention proposes a computer executable program, which, when executed, implements the method described in any one of the first aspects.

[0019] (III) Beneficial effects

[0020] The present invention groups multiple communication interactions in the recorded information based on interaction records that last for a period of time, and calculates multiple features of each group, such as obtaining multiple dimensional vectors, and splices the features of each group to generate feature data of fixed dimensions, so as to build a corresponding prediction model according to the feature data to predict the interaction trend, control whether to occupy communication resources to continue communication interaction (such as calls) according to the feature data, and / or evaluate the state of communication interaction according to the feature data, etc. Thus, feature analysis of continuous multiple interactions is achieved.

[0021] Furthermore, this analysis method, by grouping and then splicing features, can always generate feature data of a fixed number of dimensions, regardless of whether the dynamic changes of continuous interaction increase or not. This is simple and easy, and greatly reduces the difficulty of subsequent construction of prediction models, evaluation models, control processing, etc., which utilizes feature data, thereby greatly improving the efficiency of data analysis and processing.

[0022] Furthermore, by grouping, features can be calculated and concatenated more efficiently, and trends can be judged and predicted more accurately to determine whether it is necessary to control and continue to follow up on communication interactions, so as to avoid excessive communication resource occupation and reduced interaction efficiency, and improve the good experience and communication efficiency of both parties in the communication interaction.

[0023] In general, the technical solution of the present invention can realize simple, efficient and accurate feature analysis for multiple communication interactions over a period of time, determine fixed-dimensional feature data, and thus improve the effective control and control accuracy of intelligent communication control, better predict the development and change trend of communication interaction, and more accurately and efficiently evaluate the interaction status. Especially in the case of intelligent human-computer communication interaction, in a sustainable and dynamically increasing call environment, accurate and efficient control and management of communication interaction can be achieved, communication resource usage can be reduced, and user experience can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a main flow chart of an embodiment of a feature analysis method based on continuous interaction according to the present invention;

[0025] Figure 2 is a main structural block diagram of an embodiment of a feature analysis device based on continuous interaction according to the present invention;

[0026] Figure 3 is a main structural block diagram of an embodiment of an electronic device according to the present invention;

[0027] Figure 4 is a schematic diagram of the main structure of an embodiment of a more specific electronic device according to the present invention;

[0028] Figure 5 It is a main structural diagram of an embodiment of a computer-readable medium according to the present invention. DETAILED DESCRIPTION

[0029] In the process of introducing specific embodiments, the detailed description of the structure, performance, effect or other characteristics is to enable those skilled in the art to fully understand the embodiments. However, it does not exclude that those skilled in the art can implement the present invention with a technical solution that does not contain the above-mentioned structure, performance, effect or other characteristics under certain circumstances.

[0030] The flowchart in the accompanying drawings is only an exemplary process demonstration, and does not mean that the solution of the present invention must include all the contents, operations and steps in the flowchart, nor does it mean that it must be executed in the order shown in the figure. For example, some operations / steps in the flowchart can be decomposed, some operations / steps can be combined or partially combined, etc. The execution order shown in the flowchart can be changed according to actual conditions without departing from the main purpose of the present invention.

[0031] Frames in the attached figure Figure 1Generally, it refers to functional entities, which do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processing unit devices and / or microcontroller devices.

[0032] The same reference numerals in the drawings represent the same or similar elements, components or parts, and thus the repeated description of the same or similar elements, components or parts may be omitted below. It should also be understood that although the first, second, third and other attributives representing the numbers may be used herein to describe various devices, elements, components or parts, these devices, elements, components or parts should not be limited by these attributives. In other words, these attributives are only used to distinguish one from another. For example, the first device may also be called the second device, but it does not deviate from the essential technical solution of the present invention. In addition, the terms "and / or" and "and / or" refer to all combinations including any one or more of the listed items.

[0033] The following is a description of some of the terms that may be involved in the description of various embodiments of the present invention:

[0034] Calls include, but are not limited to, audio, video and / or text calls and other communication interactions achieved through wired or wireless communication methods such as the Internet, traditional telephone networks, mobile communication networks, etc. In addition, calls include, but are not limited to calls / interactions / communications in the form of audio, video and / or text through various terminals including smart terminals.

[0035] Continuous interaction includes, but is not limited to: multi-frequency / multiple / multiple communication interactions that occur over a continuous period of time, interactions that increase in frequency dynamically, and the information of each interaction will be recorded and stored in chronological order (such as logs, etc.), and its frequency / number of times, content, etc. will be recorded. Among them, a continuous period of time can be the first interaction with a target, accumulated a certain number of times in a week, a certain number of times in a month, etc., and the information generated by the accumulated interactions over any continuous period of time (including records of the time when the interaction occurred, etc.) can be obtained as needed.

[0036] Multiple calls: In telemarketing scenarios, the customer service party / sales party (human or intelligent customer service robot) makes many calls to target customers (such as customers who have consulted, customers who have registered demand information, customers obtained by other means, etc.). The target customer is the target object that hopes to become or has become a customer, that is, the object with which communication interaction (communication) is required.

[0037] The main implementation process of an embodiment of the present invention is as follows: S1, obtaining interactive content that lasts for a period of time, and grouping the interactive content that belongs to the same target in the interactive content; S2, respectively calculating N interactive features in each group of interactive content and concatenating the features of all groups, and finally generating feature data. Furthermore, predicting the interactive trend based on the feature data controls whether to continue communication, etc.

[0038] [Example 1]

[0039] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the method implementation of the present invention is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0040] Here we will combine Figure 1 The main steps of an embodiment of the method of the present invention are described in the flowchart shown in FIG. The method mainly includes:

[0041] Step S110: Acquire target interactive content, and group interactive content belonging to the same target in the interactive content.

[0042] Step S120: Calculate the features of the interactions in each group of interactive content respectively, and concatenate the features of each group to generate feature data.

[0043] Step S130: Based on the characteristic data, a judgment is made to predict the interaction development trend, control whether to continue the communication interaction, and / or evaluate the quality of the interaction.

[0044] More specifically, in step S110, the target interaction content is obtained, and the interaction content belonging to the same target in the interaction content is grouped, specifically including: obtaining all interactions within a continuous period of time (the interaction content includes various related information of the interaction) and determining multiple interactions belonging to the same target from all the interactions, and grouping the multiple interactions.

[0045] In one embodiment, obtaining the target interaction content specifically includes: selecting any continuous time period as a continuous time period for analysis according to analysis requirements, obtaining all interactions within the continuous time period, and determining multiple interactions belonging to the same target among all interactions. The continuous time period is any time period selected according to analysis requirements as a continuous time period for analysis. The interaction content includes N interactions, such as all interactions within the time period. Further, grouping the interaction content belonging to the same target in the interaction content specifically includes: dividing the multiple interactions belonging to the same target into two groups according to predetermined conditions, wherein the multiple interactions include each interaction information recorded corresponding to the multiple interactions.

[0046] Furthermore, in the selected time period, i.e., the continuous period of time for analysis, the number of the multiple interactions belonging to the same target is accumulated in chronological order and the multiple interactions are extracted; the predetermined conditions include: the accumulated number is an odd number or an even number; the interaction information includes the time of the interaction.

[0047] Further, the multiple interactions are divided into two groups according to predetermined conditions, for example: when the accumulated number is an odd number 2n+1, the first n interactions and the n+1th interaction are the first group, and the n+1th interaction to the 2n+1th interaction are the second group; when the accumulated number is an even number 2n, the first n interactions are the first group, and the n+1th interaction to the 2nth interaction are the second group; wherein n is an integer greater than or equal to 0.

[0048] In one implementation, this embodiment is applied in a telemarketing scenario, for example: the target interaction content can actually be multiple calls with a customer, and similarly, the multiple interactions are multiple calls with the same customer in multiple calls; the voice of the call is converted to text by FR / ASR to obtain text information corresponding to the call; the target is the customer in the interaction; obtaining the target interaction content can, for example, extract multiple calls with the customer from all logs over a period of time, and determine multiple interactions / multiple calls belonging to the same customer in these multiple calls based on the customer identity. Specific examples:

[0049] Extract multiple calls from all logs over a period of time, and determine multiple interactions belonging to the same customer based on the customer identity. You can first obtain multiple calls and extract multiple calls belonging to the same target. Taking the use of customer service robots in telemarketing scenarios as an example, the customer service robot communicates and visits target customers in sequence among newly registered or newly registered or other target customers obtained through other channels. For example, it obtains the registered communication information of the target customer, such as telephone number, and then calls the company's marketing number in sequence to communicate and interact to recommend company products, etc.; it can also be that new target customers automatically call the marketing number of the customer service robot to inquire about product information, etc.

[0050] In a time period, multiple calls can be made / communicated, and records of multiple calls can be kept. At intervals of a fixed time period, a pre-set time period, or an optional time period, multiple calls of one or several cycles / time periods can be obtained from the record, and then multiple calls belonging to the same target customer can be extracted based on the same telephone numbers and / or other unique identification identifiers of the communication interaction (WeChat ID, network platform registration account, unique account assigned during random access, etc.). Select a time period that needs or wants to be analyzed to obtain multiple calls from it, and extract multiple calls of the same target, accumulate the number, and analyze the call development trend of the target.

[0051] Furthermore, the grouping includes but is not limited to two groups, and can be equally divided into three groups, four groups, etc. Preferably, two groups are used.

[0052] In one implementation, multiple calls belonging to the same target are grouped according to predetermined conditions, such as grouping according to the odd and even numbers of the calls. For example:

[0053] Throughout the month of May 2020, multiple calls / conversations numbered 001, 002, 007, 009, 010, and 015 in chronological order belonged to target customer A. There were a total of 6 calls, an even number, divided into two groups, in chronological order, the first group / 001, 002, 007; the second group / 009, 010, 015.

[0054] Throughout the month of May 2020, multiple calls / conversations numbered 003, 004, 005, 006, and 008 in chronological order belonged to target customer B. There were a total of 5 calls, an odd number, and they were evenly divided into two groups. In chronological order, the first group / the first group were 003, 004, and 005; the second group / the second group were 005, 006, and 008.

[0055] It can be seen that the number of accumulated calls will be different as time goes by or different time periods are selected. The dynamically increased calls and their corresponding call feature dimensions are also changing dynamically. For feature data with non-fixed dimensions, there will be many difficulties in subsequent processing. Therefore, by grouping and calculating the N features of each group separately (taking the feature vectors, adding them and averaging them, and then taking the fixed dimension), and splicing the features of the two groups, we can observe and compare the situations of the two groups before and after, determine the changes in the features of the two groups, and judge whether the trend of the interaction is positive or negative.

[0056] In step S120, the features of the interactions in each group of interactive contents are calculated respectively, the features of each group are concatenated, and finally feature data is generated, which specifically includes:

[0057] In one embodiment, features are obtained from the text information of each interaction in each group; the features of all the interactions in each group are added together and then the average is taken; features of a preset fixed dimension are extracted from the average; the features of the fixed dimension extracted from each group are added together to generate feature data of twice the fixed dimension.

[0058] In one embodiment, the text information is text information input during interaction, text information converted from audio via FR or speech recognition during interaction, and / or text information converted from video images via character recognition or image recognition during interaction; the feature is a multidimensional feature vector; the feature vector is obtained from the text information based on word segmentation, keyword, and / or word frequency calculation. For example: the multiple interactions are multiple calls; the voice of the calls is converted into text information corresponding to the calls after FR / ASR conversion; the target is the customer in the interaction. Feature data can be directly obtained or recognized from audio or video based on the feature calculation model constructed in the actual application scenario.

[0059] In one implementation, the features of each group are calculated in each group. For example, for customer A, feature vectors are obtained from the text information of the first group 001, 002, and 007. The feature vectors obtained from these three calls are added and averaged. The 100-dimensional feature vector after the average is taken as the result of the first group, and the second group 009, 010, and 015 also obtain the second group result of 100 dimensions. The simplest splicing, such as adding, is 200 dimensions. According to the situation of the two groups before and after, the development trend of the call can be preliminarily judged, whether it is positive or negative, so as to determine whether customer A is willing to communicate, etc. Further, the feature data of 200 dimensions is used to input into various trained models for prediction (trend judgment); and / or the judgment result of positive or negative is used to determine whether to stop following up with customer A, so as not to occupy communication resources, etc.; and / or input into the evaluation model to evaluate the state of interactive communication. A high score indicates that the communication is smooth and the speech is excellent, and a low score indicates that the communication is not smooth and the speech is poor, etc.

[0060] The feature vector can be obtained from the text information by: obtaining the keywords in the text information according to the vocabulary such as the scene corpus, calculating the frequency of the keywords, sorting the keywords with high frequency to extract the corresponding feature vector; or: after word segmentation, the words that are irrelevant to the business and invalid and have no influence can be removed from the text information, the frequency of each word in the word segmentation can be calculated, the first M words can be sorted and taken as keywords, and then the corresponding feature vector can be extracted. For example, the word segmentation can be performed first, and then the TFIDF algorithm can be used for vectorization (using only the word frequency is not the best effect), and then the feature selection algorithm or feature extraction algorithm can be used to reduce the dimension (such as information entropy, information gain, etc.) to obtain the feature vector in the text information.

[0061] It can be seen that the analysis of the information recorded for each interaction in each group is actually the analysis of a single interaction such as a single call, that is, a single analysis. The features of multiple single interactions in each group are calculated separately, added up and averaged, that is, N features are calculated, the preset first L dimensions of the feature vector are taken to represent the feature vector, the L-dimensional features of the previous and next groups, preferably two groups, are spliced ​​and directly added together to generate feature data, the interaction trend changes of the previous and next groups are observed, and the spliced ​​feature data are processed in various ways required later. Such calculation and analysis are simple, easy to implement, efficient, efficient, convenient and accurate when used for subsequent processing.

[0062] In step S130, judging based on the characteristic data to predict the interaction development trend, controlling whether to continue the communication interaction, and / or evaluating the quality of the interaction specifically includes:

[0063] In one implementation, a prediction model for determining the interaction development trend may be constructed, and the feature data may be input into the trained prediction model. The interaction development trend may be determined based on the output prediction result to determine whether the interaction development trend is positive or negative.

[0064] In one implementation, the interaction development trend (such as comparison between previous and subsequent groups) is determined based on the characteristic data to control whether it is necessary to continue to occupy communication resources to interact with the target communication.

[0065] In one implementation, an evaluation model for evaluating the quality of an interaction state is constructed, the feature data is input into the trained evaluation model, and the interaction is judged to be in different scoring states according to the evaluation result.

[0066] In one implementation, trends can be analyzed by building and training a prediction model, and then concatenating feature data and inputting it into the prediction model for prediction.

[0067] For example, the audio of the call is converted into voice or the video is converted into image to generate text information. The conversion method can be through the commonly used audio recognition and image recognition methods. If the phone uses text communication, the text information is directly extracted. Further, the feature vector is obtained from the text information. For the telemarketing scenario, the feature vector includes price, product name, product number, production or sales time, delivery time, product quantity, etc. It can be extracted according to the actual application scenario or dynamically according to the text information. According to the grouping, these feature vectors are calculated and added and averaged, and the first 100 dimensions are determined as the result output of the feature vector. The feature vectors of the two groups add up to 200 dimensions, that is, feature data. The 200-dimensional feature data is used to input various existing models, prediction models, scoring models, etc. (specifically common CNN, RNN, BERT and other models) for prediction, judgment, scoring, etc. This type of model algorithm will not be described here.

[0068] It can be seen that no matter what model and demand are used later, the dynamic increase in the number of interactions and the changes in the feature dimensions caused by the changes in the interaction information, in the process of calculating N features in groups (adding and averaging fixed dimensions), the dimension / number of feature data can always be fixed, which will not interfere with the processing of feature data required later, and no more preprocessing is required. Data with unfixed dimensions is difficult to predict directly using model algorithms. At the same time, it takes into account the analysis of multiple interactions / calls, and is more objective and accurate in predicting interaction trends, whether to continue tracking communications, and whether communication resources need to be occupied. It is simple, easy and efficient to obtain reasonable feature data, which helps to improve the user experience.

[0069] Therefore, by analyzing the changes in the information of the single-channel and / or multi-channel communication interaction records generated over a period of time and comprehensively judging multiple interactions, feature data (dimensions) are generated / determined through feature analysis of grouped calculation features based on the information of the interactive communication, especially by calculating the features of the groups separately and generating feature data based on feature splicing to judge the trend, efficient and accurate feature analysis can be achieved to determine fixed-dimensional feature data, so as to improve the effective control and control accuracy of intelligent communication control, better predict the development and change trend of communication interaction, and more accurately and efficiently evaluate the state of communication. Especially during intelligent human-machine communication interaction, in a sustainable and dynamically increasing call environment, accurate and efficient control and management of communication interaction can be achieved, communication resource usage can be reduced, and user experience can be improved.

[0070] [Example 2]

[0071] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the device implementation of the present invention is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0072] Here we will combine Figure 2 The main structural block diagram of an embodiment of the method of the present invention is shown. In this embodiment, the device at least includes a grouping module B210, which is used to: obtain target interactive content and group interactive content belonging to the same target in the interactive content; a generation module B220, which is used to calculate the characteristics of the interaction in each group of interactive content, splice the characteristics of each group, and generate characteristic data; a prediction module B230, which is used to make a judgment based on the characteristic data to predict the development trend of the interaction, control whether to continue the communication interaction, and / or evaluate the quality of the interaction.

[0073] In the grouping module B210, the target interaction content is obtained, and the interaction content belonging to the same target in the interaction content is grouped specifically including: obtaining all interactions within a continuous period of time (the interaction content includes various related information of the interaction) and determining multiple interactions belonging to the same target from all the interactions, and grouping the multiple interactions.

[0074] In one embodiment, obtaining the target interaction content specifically includes: selecting any continuous time period as a continuous time period for analysis according to analysis requirements, obtaining all interactions within the continuous time period, and determining multiple interactions belonging to the same target among all interactions. The continuous time period is any time period selected according to analysis requirements as a continuous time period for analysis. The interaction content includes N interactions, such as all interactions within the time period. Further, grouping the interaction content belonging to the same target in the interaction content specifically includes: dividing the multiple interactions belonging to the same target into two groups according to predetermined conditions, wherein the multiple interactions include each interaction information recorded corresponding to the multiple interactions.

[0075] Furthermore, in the selected time period, i.e., the continuous period of time for analysis, the number of the multiple interactions belonging to the same target is accumulated in chronological order and the multiple interactions are extracted; the predetermined conditions include: the accumulated number is an odd number or an even number; the interaction information includes the time of the interaction.

[0076] Further, the multiple interactions are divided into two groups according to predetermined conditions, for example: when the accumulated number is an odd number 2n+1, the first n interactions and the n+1th interaction are the first group, and the n+1th interaction to the 2n+1th interaction are the second group; when the accumulated number is an even number 2n, the first n interactions are the first group, and the n+1th interaction to the 2nth interaction are the second group; wherein n is an integer greater than or equal to 0.

[0077] In one implementation, this embodiment is applied in a telemarketing scenario, for example: the target interaction content can actually be multiple calls with a customer, and similarly, the multiple interactions are multiple calls with the same customer in multiple calls; the voice of the call is converted to text by FR / ASR to obtain text information corresponding to the call; the target is the customer in the interaction; obtaining the target interaction content can, for example, extract multiple calls with the customer from all logs over a period of time, and determine multiple interactions / multiple calls belonging to the same customer in these multiple calls based on the customer identity. Specific examples:

[0078] Extract multiple calls from all logs over a period of time, and determine multiple interactions belonging to the same customer based on the customer identity. You can first obtain multiple calls and extract multiple calls belonging to the same target. Taking the use of customer service robots in telemarketing scenarios as an example, the customer service robot communicates and visits target customers in sequence among newly registered or newly registered or other target customers obtained through other channels. For example, it obtains the registered communication information of the target customer, such as telephone number, and then calls the company's marketing number in sequence to communicate and interact to recommend company products, etc.; it can also be that new target customers automatically call the marketing number of the customer service robot to inquire about product information, etc.

[0079] In a time period, multiple calls can be made / communicated, and records of multiple calls can be kept. At intervals of a fixed time period, a pre-set time period, or an optional time period, multiple calls of one or several cycles / time periods can be obtained from the record, and then multiple calls belonging to the same target customer can be extracted based on the same telephone numbers and / or other unique identification identifiers of the communication interaction (WeChat ID, network platform registration account, unique account assigned during random access, etc.). Select a time period that needs or wants to be analyzed to obtain multiple calls from it, and extract multiple calls of the same target, accumulate the number, and analyze the call development trend of the target.

[0080] Furthermore, the grouping includes but is not limited to two groups, and can be equally divided into three groups, four groups, etc. Preferably, two groups are used.

[0081] In one implementation, multiple calls belonging to the same target are grouped according to predetermined conditions, such as grouping according to the odd and even numbers of the calls. For example:

[0082] Throughout the month of May 2020, multiple calls / conversations numbered 001, 002, 007, 009, 010, and 015 in chronological order belonged to target customer A. There were a total of 6 calls, an even number, divided into two groups, in chronological order, the first group / 001, 002, 007; the second group / 009, 010, 015.

[0083] Throughout the month of May 2020, multiple calls / conversations numbered 003, 004, 005, 006, and 008 in chronological order belonged to target customer B. There were a total of 5 calls, an odd number, and they were evenly divided into two groups. In chronological order, the first group / the first group were 003, 004, and 005; the second group / the second group were 005, 006, and 008.

[0084] It can be seen that the number of accumulated calls will be different as time goes by or different time periods are selected. The dynamically increased calls and their corresponding call feature dimensions are also changing dynamically. For feature data with non-fixed dimensions, there will be many difficulties in subsequent processing. Therefore, by grouping and calculating the N features of each group separately (taking the feature vectors, adding them and averaging them, and then taking the fixed dimension), and splicing the features of the two groups, we can observe and compare the situations of the two groups before and after, determine the changes in the features of the two groups, and judge whether the trend of the interaction is positive or negative.

[0085] For details, please refer to the processing of step S110.

[0086] In the generation module B220, the features of the interaction in each group of interaction contents are calculated respectively, and the features of each group are concatenated to generate feature data, which specifically includes:

[0087] In one embodiment, features are obtained from the text information of each interaction in each group; the features of all the interactions in each group are added together and then the average is taken; features of a preset fixed dimension are extracted from the average; the features of the fixed dimension extracted from each group are added together to generate feature data of twice the fixed dimension.

[0088] In one embodiment, the text information is text information input during interaction, text information converted from audio via FR or speech recognition during interaction, and / or text information converted from video images via character recognition or image recognition during interaction; the feature is a multidimensional feature vector; the feature vector is obtained from the text information based on word segmentation, keyword, and / or word frequency calculation. For example: the multiple interactions are multiple calls; the voice of the calls is converted into text information corresponding to the calls after FR / ASR conversion; the target is the customer in the interaction. Feature data can be directly obtained or recognized from audio or video based on the feature calculation model constructed in the actual application scenario.

[0089] In one implementation, the features of each group are calculated in each group. For example, for customer A, feature vectors are obtained from the text information of the first group 001, 002, and 007. The feature vectors obtained from these three calls are added and averaged. The 100-dimensional feature vector after the average is taken as the result of the first group, and the second group 009, 010, and 015 also obtain the second group result of 100 dimensions. The simplest splicing, such as adding, is 200 dimensions. According to the situation of the two groups before and after, the development trend of the call can be preliminarily judged, whether it is positive or negative, so as to determine whether customer A is willing to communicate, etc. Further, the feature data of 200 dimensions is used to input into various trained models for prediction (trend judgment); and / or the judgment result of positive or negative is used to determine whether to stop following up with customer A, so as not to occupy communication resources, etc.; and / or input into the evaluation model to evaluate the state of interactive communication. A high score indicates that the communication is smooth and the speech is excellent, and a low score indicates that the communication is not smooth and the speech is poor, etc.

[0090] The feature vector can be obtained from the text information by: obtaining the keywords in the text information according to the vocabulary such as the scene corpus, calculating the frequency of the keywords, sorting the keywords with high frequency to extract the corresponding feature vector; or: after word segmentation, the words that are irrelevant to the business and invalid and have no influence can be removed from the text information, the frequency of each word in the word segmentation can be calculated, the first M words can be sorted and taken as keywords, and then the corresponding feature vector can be extracted. For example, the word segmentation can be performed first, and then the TFIDF algorithm can be used for vectorization (using only the word frequency is not the best effect), and then the feature selection algorithm or feature extraction algorithm can be used to reduce the dimension (such as information entropy, information gain, etc.) to obtain the feature vector in the text information.

[0091] It can be seen that the analysis of the information recorded for each interaction in each group is actually the analysis of a single interaction such as a single call, that is, a single analysis. The features of multiple single interactions in each group are calculated separately, added up and averaged, that is, N features are calculated, the preset first L dimensions of the feature vector are taken to represent the feature vector, the L-dimensional features of the previous and next groups, preferably two groups, are spliced ​​and directly added together to generate feature data, the interaction trend changes of the previous and next groups are observed, and the spliced ​​feature data are processed in various ways required later. Such calculation and analysis are simple, easy to implement, efficient, efficient, convenient and accurate when used for subsequent processing.

[0092] In the prediction module 230, based on the characteristic data, a judgment is made to predict the interaction development trend, control whether to continue the communication interaction, and / or evaluate the quality of the interaction, specifically including:

[0093] In one implementation, a prediction model for determining the interaction development trend may be constructed, and the feature data may be input into the trained prediction model. The interaction development trend may be determined based on the output prediction result to determine whether the interaction development trend is positive or negative.

[0094] In one implementation, the interaction development trend (such as comparison between previous and subsequent groups) is determined based on the characteristic data to control whether it is necessary to continue to occupy communication resources to interact with the target communication.

[0095] In one implementation, an evaluation model for evaluating the quality of an interaction state is constructed, the feature data is input into the trained evaluation model, and the interaction is judged to be in different scoring states according to the evaluation result.

[0096] In one implementation, trends can be analyzed by building and training a prediction model, and then concatenating feature data and inputting it into the prediction model for prediction.

[0097] For example, the audio of the call is converted into voice or the video is converted into image to generate text information. The conversion method can be through the commonly used audio recognition and image recognition methods. If the phone uses text communication, the text information is directly extracted. Further, the feature vector is obtained from the text information. For the telemarketing scenario, the feature vector includes price, product name, product number, production or sales time, delivery time, product quantity, etc. It can be extracted according to the actual application scenario or dynamically according to the text information. According to the grouping, these feature vectors are calculated and added and averaged, and the first 100 dimensions are determined as the result output of the feature vector. The feature vectors of the two groups add up to 200 dimensions, that is, feature data. The 200-dimensional feature data is used to input various existing models, prediction models, scoring models, etc. (specifically common CNN, RNN, BERT and other models) for prediction, judgment, scoring, etc. This type of model algorithm will not be described here.

[0098] It can be seen that no matter what model and demand are used later, the dynamic increase in the number of interactions and the changes in the feature dimensions caused by the changes in the interaction information, in the process of calculating N features in groups (adding and averaging fixed dimensions), the dimension / number of feature data can always be fixed, which will not interfere with the processing of feature data required later, and no more preprocessing is required. Data with unfixed dimensions is difficult to predict directly using model algorithms. At the same time, it takes into account the analysis of multiple interactions / calls, and is more objective and accurate in predicting interaction trends, whether to continue tracking communications, and whether communication resources need to be occupied. It is simple, easy and efficient to obtain reasonable feature data, which helps to improve the user experience.

[0099] Therefore, by analyzing the changes in the information of the single-channel and / or multi-channel communication interaction records generated over a period of time and comprehensively judging multiple interactions, feature data (dimensions) are generated / determined through feature analysis of grouped calculation features based on the information of the interactive communication, especially by calculating the features of the groups separately and generating feature data based on feature splicing to judge the trend, efficient and accurate feature analysis can be achieved to determine fixed-dimensional feature data, so as to improve the effective control and control accuracy of intelligent communication control, better predict the development and change trend of communication interaction, and more accurately and efficiently evaluate the state of communication. Especially during intelligent human-machine communication interaction, in a sustainable and dynamically increasing call environment, accurate and efficient control and management of communication interaction can be achieved, communication resource usage can be reduced, and user experience can be improved.

[0100] [Example 3]

[0101] The following describes an overall application scenario and combines Examples 1 and 2 to further illustrate the implementation process of the present invention:

[0102] In the telemarketing scenario, according to the target customer list, multiple calls are made to a target customer A in a week, for example, one call per day, a total of seven calls, assuming the call numbers are A001, A002, ... A007. Select calls made to customer A for five consecutive working days in a week, assuming that the number is 001 to 005, from Monday to Friday, for an odd number of multiple calls, preferably divided into two groups: the first group has 001, 002, 003, and the second group has 003, 004, 005.

[0103] Using the existing NLP semantic analysis technology, based on the text conversion of the call, the call speech is recognized by ASR as the corresponding text, and then analyzed and calculated. In this way, the following features of each call can be calculated separately: whether it contains negative emotions, whether the user has made a strong refusal, whether the user wants to know the detailed information of the course, etc. The feature value of each dimension can be represented by 0 or 1. In this way, N-dimensional 0 / 1 features are obtained after calculation. The three multidimensional features of each group are added together and averaged to calculate the average value of the multidimensional features of the group (the first group or the second group). Each group takes the first N dimensions, that is, the number of dimensions (fixed length). Then the N-dimensional features of the two groups are spliced. The splicing method can be addition or direct splicing. For example, if the feature values ​​of the two segments are 110 and 011 respectively, 110011 is formed after splicing. This feature concatenation method may have slight differences when applied to different downstream prediction models. For example, if the downstream is a time series neural network such as RNN or LSTM, a separator needs to be added between each feature value during concatenation, such as 110[sep]011.

[0104] Assume that the first N-dimensional features of the two groups are taken separately, and through feature concatenation, 2*N-dimensional 0 / 1 features are generated for each target customer. At this time, the judgment of positive and negative is transformed into a classification task.

[0105] For example, based on the historical order data of telemarketing, we can select M historical target customers and calculate a 2*N-dimensional 0 / 1 feature for each target customer based on the historical call records. If the target customer finally converts, the label is 1, otherwise the label is 0.<feature,label> The binary group constitutes the training set for the classification task, which can be trained using mature classification models such as SVM and GBDT. The trained classification model is then used to predict the conversion intention of the new online target customer A based on its 2N-dimensional features (for example, predict its conversion trend) to determine whether it is positive or negative, that is, to determine whether it is possible to become an actual customer. If there is a possibility of conversion, continue to follow up, otherwise give up or delay the follow-up to avoid occupying too many communication resources, etc. The features can also be used to evaluate the quality of the call status through the model, etc.

[0106] [Example 4]

[0107] Figure 3 It is a structural schematic block diagram of an electronic device according to an embodiment of the present invention, the electronic device includes a processor and a memory, the memory is used to store a computer executable program, when the computer program is executed by the processor, the processor executes the embodiment steps of the method involved in the aforementioned embodiments 1 and 3.

[0108] like Figure 3 As shown, the electronic device is presented in the form of a general-purpose computing device. The processor may be one or more and work in coordination. The present invention does not exclude distributed processing, that is, the processor may be dispersed in different physical devices. The electronic device of the present invention is not limited to a single entity, but may also be the sum of multiple physical devices.

[0109] The memory stores a computer executable program, usually a machine-readable code, which can be executed by the processor to enable the electronic device to perform the method of the present invention, or at least part of the steps in the method.

[0110] The memory includes a volatile memory, such as a random access memory unit (RAM) and / or a cache memory unit, and may also be a non-volatile memory, such as a read-only memory unit (ROM).

[0111] Optionally, in this embodiment, the electronic device further includes an I / O interface, which is used for the electronic device to exchange data with external devices. The I / O interface can represent one or more of several types of bus structures, including a storage unit bus or a storage unit controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of a variety of bus structures.

[0112] More specifically, see Figure 4 1 is a block diagram of a more specific example of the electronic device described in the embodiment shown. The electronic device 200 of the exemplary embodiment is in the form of a general data processing device. The components of the electronic device 200 may include, but are not limited to: at least one processing unit 210, at least one storage unit 220, a bus 230 connecting different system components (including the storage unit 220 and the processing unit 210), a display unit 240, etc.

[0113] The storage unit 220 stores a computer-readable program, which may be a source program or a code of a read-only program. The program may be executed by the processing unit 210, so that the processing unit 210 performs the steps of various embodiments of the present invention. For example, the processing unit 210 may perform the various steps of the methods of the aforementioned embodiments 2 to 5.

[0114] The storage unit 220 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 2201 and / or a cache storage unit 2202, and may further include a read-only storage unit (ROM) 2203. The storage unit 220 may also include a program / utility 2204 having a set (at least one) of program modules 2205, such program modules 2205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include the implementation of a network environment.

[0115] Bus 230 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0116] The electronic device 200 may also communicate with one or more external devices 300 (e.g., keyboards, displays, network devices, Bluetooth devices, etc.) so that a user can interact with the electronic device 200 via these external devices 300, and / or so that the electronic device 200 can communicate with one or more other data processing devices (e.g., routers, modems, etc.). Such communication may be performed through an input / output (I / O) interface 250, and may also be performed through a network adapter 260 with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet). The network adapter 260 may communicate with other modules of the electronic device 200 via the bus 230. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in the electronic device 200, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0117] It should be understood that Figure 3 , 4 The electronic device shown is only an example of the present invention, and the electronic device of the present invention may also include elements or components not shown in the above examples. For example, some electronic devices also include display units such as display screens, and some electronic devices also include human-computer interaction elements such as buttons, keyboards, etc. As long as the electronic device can execute the computer-readable program in the memory to implement the method of the present invention or at least part of the steps of the method, it can be considered as an electronic device covered by the present invention.

[0118] [Example 5]

[0119] Figure 5 Schematic diagram of a computer readable recording medium according to an embodiment of the present invention. Figure 5 As shown, a computer executable program is stored in a computer-readable recording medium, and when the computer executable program is executed, the above-mentioned method for providing a tutoring strategy based on a question of the present invention is implemented. The computer-readable storage medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a 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. The readable storage medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in combination with an instruction execution system, device, or device. The program code contained on the readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0120] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0121] Through the above description of the implementation mode, it is easy for those skilled in the art to understand that the present invention can be implemented by hardware capable of executing a specific computer program, such as the system of the present invention, and the electronic processing unit, server, client, mobile phone, control unit, processor, etc. contained in the system. The present invention can also be implemented by a vehicle comprising at least a part of the above system or components. The present invention can also be implemented by computer software that executes the method of the present invention, such as control software executed by a microprocessor, an electronic control unit, a client, a server, etc. on the locomotive side. However, it should be noted that the computer software that executes the method of the present invention is not limited to being executed by one or a specific hardware entity, and it can also be implemented in a distributed manner by unspecified specific hardware, such as some method steps executed by the computer program can be executed on the locomotive side, and another part can be executed in a mobile terminal or a smart helmet, etc. For computer software, the software product can be stored in a computer-readable storage medium (which can be a CD-ROM, a U disk, a mobile hard disk, etc.), or it can be distributed and stored on the network, as long as it enables the electronic device to execute the method according to the present invention.

[0122] Through the above description of the implementation modes, those skilled in the art can easily understand that the exemplary embodiments described in the present invention can be implemented by software, or by combining software with necessary hardware.

[0123] The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the present invention is not inherently related to any specific computer, virtual device or electronic device, and various general devices can also implement the present invention. The above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A feature analysis method based on continuous interaction, characterized in that: include: Acquiring target interaction content, including: selecting any continuous time period as a continuous time period for analysis according to analysis requirements, acquiring all interactions within the continuous time period, and determining multiple interactions belonging to the same target among all the interactions; Grouping the interactive contents belonging to the same target in the interactive contents, including: dividing the multiple interactions belonging to the same target into two groups according to a predetermined condition, wherein the multiple interactions include each interaction information recorded corresponding to the multiple interactions; Calculating the features of the interactions in each group of interactive contents respectively, including: obtaining features from text information of each interaction in each group, adding the features of all interactions in each group and then taking an average value, and extracting features of a preset fixed dimension from the average value; wherein the text information is text information input during the interaction, text information converted from audio through FR or speech recognition during the interaction, and / or text information converted from video images through character recognition or image recognition during the interaction; The features of each group are concatenated to generate feature data, including: adding the features of fixed dimension extracted from each group to generate feature data of twice the fixed dimension; wherein the feature is a multidimensional feature vector, and the feature vector is obtained from the text information based on word segmentation, keywords, and / or word frequency calculation.

2. The method according to claim 1, characterized in that Also includes: Accumulating the number of the multiple interactions belonging to the same target in chronological order within the continuous period of time, and extracting the multiple interactions; The predetermined condition includes: the accumulated number of times is an odd number or an even number; The interaction information includes the time of the interaction.

3. The method according to claim 2, characterized in that The multiple interactions belonging to the same target are divided into two groups according to predetermined conditions, specifically including: When the accumulated number is an odd number 2n+1, the first n interactions and the n+1th interaction are the first group, and the n+1th interaction to the 2n+1th interaction are the second group; When the accumulated number is an even number 2n, the first n interactions are the first group, and the n+1th to 2nth interactions are the second group; Wherein, n is an integer greater than or equal to 0.

4. The method according to claim 1, characterized in that The feature vector is obtained from the text information according to word segmentation, keywords, and / or word frequency calculation, including: obtaining keywords in the text information according to the scene corpus vocabulary, calculating the word frequency of the keywords, sorting the keywords with high word frequency to extract the corresponding feature vector; or, removing words that are irrelevant to the business and invalid and have no impact in the text information after word segmentation, calculating the word frequency of each word in the word segmentation, sorting and taking the first M words as keywords, and then extracting the corresponding feature vector.

5. The method according to any one of claims 1 to 4, characterized in that Also includes: The interactive content is multiple calls; The voice of the call is converted into text to obtain text information corresponding to the call; The target is the customer in the interaction; Acquiring the target interaction content also includes: acquiring multiple calls with the customer in all logs over a period of time, and determining the multiple calls belonging to the same customer according to the customer identity identifier.

6. The method according to claim 1, characterized in that Also includes: Constructing a prediction model for determining the interactive development trend, and inputting the feature data into the trained prediction model; Judging whether the interactive development trend is positive or negative according to the output prediction result, so as to determine the interactive development trend; and / or, Determine the interaction development trend according to the characteristic data, and control whether it is necessary to continue to occupy communication resources and interact with the target communication; and / or, An evaluation model for evaluating the quality of the interaction state is constructed, the feature data is input into the trained evaluation model, and the interaction is judged to be in different scoring states according to the evaluation result.

7. A feature analysis device based on continuous interaction, characterized in that: include: A grouping module is used to group the acquired target interaction content and the interaction content belonging to the same target in the interaction content, including: selecting any continuous time period as a continuous time period for analysis according to analysis requirements, acquiring all interactions within the continuous time period, and determining multiple interactions belonging to the same target in all interactions; and dividing the multiple interactions belonging to the same target into two groups according to predetermined conditions, wherein the multiple interactions include each interaction information recorded corresponding to the multiple interactions; A generation module is used to respectively calculate the features of the interactions in each group of interactive content, concatenate the features of each group, and generate feature data, including: obtaining features from the text information of each interaction in each group, adding the features of all interactions in each group and then taking an average value, and extracting features of a preset fixed dimension from the average value; wherein the text information is text information input during the interaction, text information converted from audio via FR or speech recognition during the interaction, and / or text information converted from video images via character recognition or image recognition during the interaction; and, adding the features of the fixed dimension extracted from each group to generate feature data of twice the fixed dimension; wherein the feature is a multidimensional feature vector, and the feature vector is obtained from the text information based on word segmentation, keyword, and / or word frequency calculation.

8. A computer-readable medium storing a computer-executable program, wherein when the computer-executable program is executed, the method according to any one of claims 1 to 6 is implemented.

Images