Customer follow-up information processing method and device and electronic equipment

Abstract

The invention provides a client follow-up information processing method and device and electronic equipment. The method comprises the steps that client follow-up information to be processed currently in a client management system is acquired; performing sentiment analysis prediction on the customer follow-up information through a preset sentiment analysis model, and outputting a sentiment classification tag corresponding to the customer follow-up information; determining a target customer index type corresponding to the customer follow-up information based on the emotion classification labels corresponding to the customer follow-up information; based on the sentiment classification label and the target customer index type, counting the customer follow-up information into existing billboard data; the billboard data comprises statistical data under a single label dimension and / or a combination of a plurality of label dimensions, and after an event meeting a decision triggering condition is monitored in the billboard data, decision reminding is carried out on salesmen. In the scheme of the invention, customer judgment standards are unified, automatic judgment, analysis and decision reminding can be realized, the transaction period can be shortened, the conversion rate can be improved, and the customer maintenance cost can be reduced.

Description

Technical Field

[0001] The present application relates to the field of software technology, and in particular to a method, device and electronic device for processing customer follow-up information. Background Art

[0002] In the customer management system, the content of sales follow-up is unstructured data in the form of text and voice. It is difficult to analyze which key customers are and how much customers recognize the platform. Asking sales to mark customer categories not only consumes energy, but also because each salesperson has different levels of understanding and subjective awareness, the labeling standards for sales marking are inconsistent, which greatly interferes with data analysis. In order to reduce interference, sales training is required, which is very costly. At the same time, the mobility of sales leads to the superposition of training costs. Summary of the invention

[0003] The purpose of this application is to provide a method, device and electronic device for processing customer follow-up information, which automatically identifies and classifies the customer follow-up information to be processed through a preset trained sentiment analysis model, determines the target customer indicator type based on the sentiment classification label, and updates it to the dashboard data, and then monitors the triggering decision events in the dashboard data to provide better decision reminders for sales personnel. In this method, the customer evaluation criteria are unified and can automatically judge, analyze and remind decisions, which can shorten the transaction cycle, improve conversion rate and reduce customer maintenance costs. .

[0004] In a first aspect, the present application provides a method for processing customer follow-up information, the method comprising: obtaining customer follow-up information currently to be processed in a customer management system; performing sentiment analysis and prediction on the customer follow-up information through a preset sentiment analysis model, and outputting a sentiment classification label corresponding to the customer follow-up information; the sentiment classification label represents the degree of the customer's language intention; based on the sentiment classification labels corresponding to the customer follow-up information, determining the target customer indicator type corresponding to the customer follow-up information; based on the sentiment classification labels and target customer indicator types corresponding to the customer follow-up information, counting the customer follow-up information into existing dashboard data; the dashboard data includes statistical data under a single label dimension and / or a combination of multiple label dimensions, and after monitoring an event that meets a decision trigger condition in the dashboard data, providing a decision reminder to the salesperson.

[0005] Furthermore, the above-mentioned customer follow-up information includes: text communication information between sales personnel and customers or text information converted from voice communication information; before the step of determining the target customer indicator type corresponding to the customer follow-up information based on the sentiment classification labels corresponding to the customer follow-up information, the method also includes: if customer behavior data corresponding to the customer follow-up information is obtained; predict the customer behavior data through a preset behavior analysis model to obtain a behavior classification label corresponding to the customer behavior data; the behavior classification label represents the degree of user behavior intention; judge whether the behavior classification label and the sentiment classification label are consistent; if inconsistent, correct the sentiment classification label with the behavior classification label; wherein the types corresponding to the sentiment classification label and the behavior classification label include: distrust, general, perfunctory, interested, high intention, disgust, competitor fan, considering cooperation or already cooperating.

[0006] Furthermore, the training process of the above sentiment analysis model is as follows: obtain a business sample data set; the business sample data in the business sample data set includes: customer follow-up information and corresponding sentiment classification labels; apply the business sample data in the business sample data set to iteratively train the preset classification model based on the prediction results to obtain a sentiment analysis model.

[0007] Furthermore, the business sample data in the above-mentioned application business sample data set is used to iteratively train the preset classification model based on the prediction result to correct it, and the step of obtaining the sentiment analysis model includes: taking each business sample data as the current sample data, and executing the following prediction steps: outputting the prediction result corresponding to the current sample data through the preset classification model; if the prediction result corresponding to the current sample data does not meet the expectation, performing data expansion and optimization on the current sample data to obtain new sample data; taking the new sample data as the current sample data again, and continuing to execute the prediction steps until the prediction accuracy of the preset classification model reaches the preset value; obtaining the target sample data that can no longer be optimized, correcting the prediction result of the target sample data based on the preset correction rule, and continuing to iteratively train the preset classification model until the prediction accuracy of the model meets the business requirements, and obtaining the sentiment analysis model.

[0008] Furthermore, the above-mentioned step of correcting the prediction result of the target sample data based on the preset correction rule includes: matching the prediction result corresponding to the target sample data with multiple preset rules; if the match is successful, using the prediction value corresponding to the hit rule as the corrected prediction result corresponding to the target sample data.

[0009] Furthermore, the above-mentioned step of matching the prediction results corresponding to the target sample data with multiple preset rules includes: for the first target sample data whose number of model prediction errors exceeds a threshold, performing similarity matching on the prediction results of the first target sample data through a semantic similarity model; defining multiple semantically similar sentences and corresponding prediction values ​​in the semantic similarity model; for the second target sample data whose model prediction errors are of a fixed type, using a regular expression matching method to perform rule matching on the prediction results of the second target sample data.

[0010] Furthermore, the above-mentioned dashboard data is pre-configured with multiple customer indicator types; each customer indicator type corresponds to a target label dimension, or a target dimension combination containing multiple label dimensions; based on the sentiment classification labels corresponding to the customer follow-up information, the step of determining the target customer indicator type corresponding to the customer follow-up information includes: according to the correspondence between the customer indicator type and the label, from multiple customer indicator types, searching for the type to which the sentiment classification label belongs; and determining the found type as the target customer indicator type corresponding to the customer follow-up information.

[0011] Furthermore, the above-mentioned step of reminding the sales staff of making decisions after monitoring the events that meet the decision trigger conditions in the dashboard data includes: monitoring the events that meet the decision trigger conditions in the following ways: judging whether there is data exceeding the warning value in the changing trend of the statistical data under a single label dimension and / or the statistical data under the customer indicator type; judging whether the type and number of sentiment classification labels of a single customer within a specified time period meet the preset trigger conditions; generating corresponding decision reminder information according to the event type, and sending it to the sales staff's terminal.

[0012] In a second aspect, the present application also provides a device for processing customer follow-up information, the device comprising: an information acquisition module, used to acquire customer follow-up information currently to be processed in a customer management system; a prediction module, used to perform sentiment analysis and prediction on customer follow-up information through a preset sentiment analysis model, and output a sentiment classification label corresponding to the customer follow-up information; the sentiment classification label represents the degree of the customer's language intention; an indicator type determination module, used to determine the target customer indicator type corresponding to the customer follow-up information based on the sentiment classification labels corresponding to the customer follow-up information; a data update module, used to count the customer follow-up information into existing dashboard data based on the sentiment classification labels and target customer indicator types corresponding to the customer follow-up information; the dashboard data includes statistical data under a single label dimension and / or a combination of multiple label dimensions, and a decision reminder module, used to provide decision reminders to sales personnel after monitoring events that meet decision trigger conditions in the dashboard data.

[0013] In a third aspect, the present application further provides an electronic device, comprising a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the method described in the first aspect above.

[0014] In the customer follow-up information processing method, device and electronic device provided by the present application, first, the customer follow-up information currently to be processed in the customer management system is obtained; sentiment analysis and prediction are performed on the customer follow-up information through a preset sentiment analysis model, and the sentiment classification label corresponding to the customer follow-up information is output; the sentiment classification label represents the degree of the customer's language intention; based on the sentiment classification labels corresponding to the customer follow-up information, the target customer indicator type corresponding to the customer follow-up information is determined; based on the sentiment classification labels and target customer indicator types corresponding to the customer follow-up information, the customer follow-up information is counted in the existing dashboard data; the dashboard data includes statistical data under a single label dimension and / or a combination of multiple label dimensions, and after an event that meets the decision trigger condition is monitored in the dashboard data, a decision reminder is given to the salesperson. This method of processing customer follow-up information uses a preset and trained sentiment analysis model to automatically identify and classify the customer follow-up information to be processed, determines the target customer indicator type based on the sentiment classification label, and updates it to the dashboard data, and then monitors the triggering decision events in the dashboard data to provide better decision reminders for sales personnel. In this method, the customer evaluation criteria are unified and can automatically judge, analyze, and make decision reminders, which can shorten the transaction cycle, improve conversion rate, and reduce customer maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the specific implementation methods of the present application or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0016] Figure 1 A flowchart of a method for processing customer follow-up information provided in an embodiment of the present application;

[0017] Figure 2 A flowchart of a behavior analysis process provided in an embodiment of the present application;

[0018] Figure 3 A schematic diagram of a process of processing customer follow-up information provided in an embodiment of the present application;

[0019] Figure 4 A structural block diagram of a device for processing customer follow-up information provided in an embodiment of the present application;

[0020] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The technical solution of the present application will be described clearly and completely below in conjunction with the embodiments. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0022] In the existing customer management system, sales staff usually classify customers by manually labeling them to achieve customer maintenance and management. However, due to the inconsistency between the subjective consciousness of each sales staff and the standard of understanding customers, confusion often occurs, which in turn causes fluctuations in the subsequent data analysis; if you want to unify the evaluation standards for customers, you need to train sales staff a lot, which leads to higher costs for manual training and customer maintenance, and the classification results are still not accurate enough.

[0023] Based on this, the embodiment of the present application provides a method, device and electronic device for processing customer follow-up information, which realizes automatic judgment and unified customer evaluation standards through a preset and trained sentiment analysis model, and can shorten the transaction cycle, improve conversion rate and reduce customer maintenance costs through analysis and decision reminders based on classification labels. To facilitate understanding of this embodiment, a method for processing customer follow-up information disclosed in the embodiment of the present application is first introduced in detail.

[0024] Figure 1 A flowchart of a method for processing customer follow-up information provided in an embodiment of the present application, the method specifically comprises the following steps:

[0025] Step S102, obtaining the customer follow-up information currently to be processed in the customer management system; the customer follow-up information includes: text communication information between the salesperson and the customer or text information converted from voice communication information;

[0026] Step S104, performing sentiment analysis and prediction on the customer follow-up information through a preset sentiment analysis model, and outputting a sentiment classification label corresponding to the customer follow-up information; the sentiment classification label represents the degree of the customer's language intention;

[0027] The above sentiment analysis model is obtained by iteratively training the preset classification model based on the business sample data in the business sample data set and correcting it based on the prediction results; the business sample data includes: customer follow-up information and corresponding sentiment classification labels; the correction of the prediction results is completed based on the preset correction rules.

[0028] In one embodiment, the sentiment analysis model is a model implemented based on natural language processing technology, which can automatically identify text-based customer follow-up information. At the same time, for voice-based customer follow-up information, the voice can be automatically converted into text before recognition.

[0029] After the sentiment analysis model performs sentiment analysis and prediction on customer follow-up information, it obtains the sentiment classification labels corresponding to the customer follow-up information, such as "distrust", "average", "perfunctory", "interested", "high intention", "disgusted", "competitor fan", "considering cooperation" or "already cooperating", etc.

[0030] Step S106, determining the target customer indicator type corresponding to the customer follow-up information based on the sentiment classification labels corresponding to the customer follow-up information respectively;

[0031] A variety of different customer indicator types are pre-set in the system. Usually, each customer indicator type corresponds to a target label dimension, or a target dimension combination containing multiple label dimensions. In specific implementation, the type to which the sentiment classification label belongs can be found from a variety of customer indicator types based on the correspondence between the customer indicator type and the label. The found type is determined as the target customer indicator type corresponding to the customer follow-up information.

[0032] Step S108, based on the sentiment classification labels and target customer indicator types respectively corresponding to the customer follow-up information, the customer follow-up information is counted into the existing dashboard data; the dashboard data includes statistical data under a single label dimension and / or a combination of multiple label dimensions;

[0033] For example, the dashboard data includes the statistical data under the "distrust" label, the statistical data volume for each day within a specified period of time, for example, 10 for the day, and the same for other labels. The dashboard data can include a variety of data analysis content, such as a data change trend graph under the same label, a data change comparison graph under different labels, etc. In addition, the dashboard data can also include: data change graphs under different customer indicator types, etc.

[0034] The Kanban data is displayed graphically and categorized, which is convenient for managers or sales to view. For example, taking the label "interested", the Kanban data shows a line chart for one week, such as the daily data volume for a week is 1, 3, 5, 0, 0, 0 respectively; it is easy to see the fluctuation through the chart, and it is also easy to find out if there is a problem. For example, the numbers in the past few days are similar, and suddenly the data is too high or too low, the decision-making system will trigger a reminder, and the sales staff will find out the problem through the chart, and then optimize the system or adjust the customer maintenance strategy.

[0035] Step S110, after an event that meets the decision triggering condition is detected in the dashboard data, a decision reminder is given to the salesperson.

[0036] There are many specific monitoring methods, such as monitoring the changes in statistical data of a single label dimension, or monitoring the changes in statistical data under a combination of label dimensions, or monitoring the label situation of a single customer, etc., which are not specifically limited here.

[0037] The processing method of customer follow-up information provided by the embodiment of the present application automatically identifies and classifies the customer follow-up information to be processed through a preset trained sentiment analysis model, determines the target customer indicator type based on the sentiment classification label, and updates it to the dashboard data, and then monitors the triggering decision events in the dashboard data to remind sales staff to make better decisions. In this method, the customer follow-up information is sentimentally classified through the model, the customer evaluation standard is unified, the manual intervention is reduced to improve efficiency, and the training cost of sales staff is reduced; the unified structured data labeled by the model brings convenience in data statistics, and it is easy to make data analysis dashboards and indicator dashboards; when the standard is changed, only a new model needs to be retrained, and the historical data is retrospectively adjusted to adjust all data to the latest standard, which can save a lot of labor costs. After the model is labeled, data analysis and statistics can be performed on the sales follow-up process, which is convenient for sales to make decisions and adjust different sales strategies, focus on following up the key customers analyzed, shorten the transaction cycle, and improve the conversion rate; promote the functions and platform usage scenarios and skills for non-key customers, and cultivate customer awareness to improve their recognition of platform products.

[0038] The embodiment of the present application also provides another method for processing customer follow-up information, which is implemented on the basis of the above embodiment; this embodiment focuses on describing the model training process and application process, data analysis and display process, and decision reminder process.

[0039] See also Figure 2 The entire solution flow chart shown in the figure shows that the training process of the above sentiment analysis model is as follows:

[0040] (1) Obtaining a business sample data set; the business sample data in the business sample data set includes: customer follow-up information and corresponding sentiment classification labels;

[0041] In specific implementation, the text content of a complete customer follow-up record (if it is voice, the voice is converted into text) is saved in the follow_up_content field in the sample database. Then the sample data is manually labeled (such as distrust, general, perfunctory, interested, high intention, disgust, competitor fan, etc.) and saved in the label field in the sample database. Alternatively, specific keywords can be used to classify and label the follow-up record content.

[0042] (2) Using the business sample data in the business sample data set, the preset classification model is iteratively trained and corrected based on the prediction results to obtain a sentiment analysis model.

[0043] During specific training, the sample data is divided into three parts: 70% training data, 20% validation data, and 10% test data. The sample data is divided into three parts according to this ratio. The preset classification model StructBERT sentiment classification-Chinese-seven categories-base is trained using the above sample data. Of course, the above classification model can also use other classification basic models.

[0044] The steps of applying the business sample data in the business sample data set to iteratively train the preset classification model based on the prediction results to obtain the sentiment analysis model include:

[0045] Take each business sample data as the current sample data and perform the following prediction steps:

[0046] 1) Output the prediction result corresponding to the current sample data through the preset classification model;

[0047] 2) If the prediction result corresponding to the current sample data does not meet expectations, the current sample data is expanded and optimized to obtain new sample data;

[0048] 3) The new sample data is used as the current sample data again, and the prediction step is continued until the prediction accuracy of the preset classification model reaches a preset value;

[0049] 4) Obtain target sample data that cannot be further optimized, correct the prediction results of the target sample data based on the preset correction rules, and continue to iteratively train the preset classification model until the prediction accuracy of the model meets the business requirements, thereby obtaining a sentiment analysis model.

[0050] In the specific implementation, the prediction results corresponding to the target sample data are first matched with multiple preset rules; two matching methods are provided here. For the first target sample data whose number of model prediction errors exceeds a threshold, the prediction results of the first target sample data are similarly matched through a semantic similarity model; multiple semantically similar sentences and corresponding prediction values ​​are defined in the semantic similarity model; for the second target sample data whose model prediction errors are of a fixed type, the prediction results of the second target sample data are rule matched using a regular expression matching method.

[0051] If the match is successful, the prediction value corresponding to the hit rule is used as the corrected prediction result corresponding to the target sample data.

[0052] For example: the specific content of the customer follow-up information is "not needed", and the model predicts "interested". The actual result you want to predict is "average", but this prediction result cannot be corrected through infusion of corpus and training. At this time, the rules are used to match the follow-up content as "not needed", and the processed result is given directly after processing through the rules, without sending it to the model for prediction.

[0053] In the embodiment of the present application, natural language processing technology is used to allow the model to understand the customer's emotions and recognition of the platform. The standards are unified and can be automatically judged. If the standards change, the model only needs to be retrained to obtain the latest standard data. The customer follow-up information can be predicted by the above-trained sentiment analysis model, and the customer follow-up information can be labeled.

[0054] In another preferred embodiment, not only the language or voice information of the above-mentioned customer is considered, but also the customer's behavior data is comprehensively considered. Specifically, before the step of determining the target customer indicator type corresponding to the customer follow-up information based on the sentiment classification labels corresponding to the customer follow-up information, the method also includes the following steps: Figure 2 As shown:

[0055] Step S202: if the customer behavior data corresponding to the customer follow-up information is obtained, the customer behavior data is predicted by a preset behavior analysis model to obtain a behavior classification label corresponding to the customer behavior data; the behavior classification label represents the degree of the user's behavior intention;

[0056] Customer behavior data includes: customer access and product usage behaviors, including clicks, favorites, assignments, searches, and other dimensions;

[0057] Step S204, determining whether the behavior classification label and the emotion classification label are consistent;

[0058] Step S206, if there is inconsistency, the sentiment classification label is corrected with the behavior classification label; wherein the types corresponding to the sentiment classification label and the behavior classification model include: distrust, general, perfunctory, interested, high intention, disgust, competitor fan, considering cooperation or already cooperated.

[0059] In the embodiment of the present application, the behavior prediction result is a key indicator. Since many customers do not have behavior data, the language or voice follow-up information is first analyzed by the sentiment analysis model. However, if the behavior data of the same user can be obtained, the behavior prediction result shall prevail. That is, when the behavior prediction label and the sentiment prediction label are inconsistent, the behavior classification label shall be replaced by the sentiment classification label.

[0060] The dashboard data can be analyzed and visualized through the data analysis tool "FineBI". The data display and interaction modules can also use other commercial software such as Microsoft's "PowerBI" or the open source "ApacheSuperset".

[0061] In specific implementation, the CRM system records sales follow-up information through a third-party outbound call system. After a follow-up call, the third-party outbound call system will push the recording file and call duration and other information to the synchronization interface of the CRM system. The synchronization interface receives the push information and downloads it, and converts the recording into text by parsing the recording file. Then, the user sentiment is analyzed and labeled through the model, and a record of the corresponding dimension is inserted into the dimension fact table of the data analysis. For example, if the analysis result is "interested", then a record of the corresponding dimension will be inserted into the dimension fact table under this label dimension. For example, the data volume before the insertion is 3, and the data volume after adding a new one is 4. In this way, the statistical changes of customer sentiment classification can be monitored through the analysis system. Once the set threshold is reached, the strategy of the decision system will be triggered.

[0062] In addition to the statistical data under the single label dimension, the dashboard data also includes the statistical data under different customer indicator types. The statistical data under different customer indicator types are obtained in the following way: according to the label dimension, the statistical data under the single dimension are classified and combined to generate customer indicator type data. The specific process is as follows:

[0063] The above sentiment classification labels and behavior classification labels include: distrust, general, perfunctory, interested, high intention, disgust, competitor fan, considering cooperation or already cooperated, etc.

[0064] 1) Combine customer data under the labels of interested, considering cooperation, and already cooperating into high-intent customer indicator type data; 2) Combine customer data under the labels of general and competitor fan into low-intent customer indicator type data; 3) Combine customer data under the labels of distrust, general, and perfunctory into low-intent customer indicator type data to be mined; 4) Combine customer data under the labels of distrust and disgust into customer indicator type data to be stimulated and cultivated; 5) Use customer data under the label of interest as key development and marketing customer indicator type data; 6) Use customer data under the label of disgust as key analysis and review customer indicator type data; 7) Use customer data under the label of high intention as customer indicator type data for continuous product improvement.

[0065] Finally, monitoring is performed based on the statistical data in the dashboard data to make decision reminders. There are many possible implementation options. For example, monitoring events that meet the decision trigger conditions in the following ways: judging whether data exceeding the warning value appears in the statistical data under a single label dimension and / or the changing trend of the statistical data under the customer indicator type; judging whether the type and number of sentiment classification labels of a single customer within a specified time period meet the preset trigger conditions; generating corresponding decision reminder information based on the event type and sending it to the salesperson's terminal.

[0066] For the statistical data under the customer indicator type, you can also pre-configure the corresponding decision reminder information for each customer indicator type, and provide decision reminders to sales personnel through search. For the label situation of a single customer, you can also combine the customer's current follow-up status and the corresponding customer indicator type to automatically generate decision reminder information.

[0067] Through the above methods, we can follow up on key customers who have made decisions and cultivate platform value for non-key customers. Figure 3 The entire process of processing customer follow-up information is shown, which is consistent with the above content and will not be repeated here.

[0068] Here are a few classic case scenarios:

[0069] Classic scenario 1: During the follow-up process, the sales staff forgets to follow up in time with a high-intent customer predicted by the model because they are busy with other customers or developing new customers. Then, when the decision system discovers this situation, it will send a message to remind the sales staff to follow up in time. For example, this customer has been followed up many times and has been judged to have high intention by the model. The follow-up has now reached the stage where information needs to be sent, but the sales staff forgot about it due to busy work. However, the decision system finds that this customer has high intention but has not followed up for a long time, so it triggers the decision to send a message reminder. Then the sales staff sees the reminder and sends the materials and quotation to the customer in time. After seeing the materials and quotation, the customer has a deeper experience of the system and decides to make a deal.

[0070] Classic scenario 2: When the salesperson follows up with the customer, the customer always or often verbally refuses to communicate in depth, or even shows bad emotions. At this time, the model determines that the customer is a low-intent customer. After multiple follow-ups, the customer is still the same. When it is judged as a low-intent customer five times, the decision system is triggered to send a reminder to "throw it into the high seas". After receiving this reminder, the salesperson combines his actual follow-up situation and his feelings about the customer to determine that he does meet the characteristics of a non-high-quality customer and then throws him into the high seas. However, after a period of time, the behavioral analysis model of the behavioral system detects that this customer often uses the functions of the company's products, triggering the decision system to send a reminder to "recover the customer". The salesperson sees this information and finds that this customer has a strong demand for the company's products, but may just not want to be disturbed. Then, the salesperson changes the sales strategy and stops calling him. Instead, he pushes some product highlights and differentiation of competing products, as well as a full range of product function introductions and success stories through text messages, WeChat or APP. Finally, the customer understands clearly and feels that the company's products are more in line with their needs, so the deal is finally made.

[0071] Classic scenario three: Due to the problem of sales staff capacity, the number of customers each sales staff has is limited, so some customers have to be abandoned and go to the high seas. Then, when customers with high language intention and high behavioral intention trigger the indicators set by the decision system, the decision system will trigger a specific reminder to notify the sales staff to develop this customer. The advantage of such intelligent processing is that customers can be found more accurately and timely without having to look for a needle in a haystack.

[0072] The method for processing customer follow-up information provided in the embodiment of the present application makes the sales follow-up process statistically analyzable, predictable and traceable through structured customer follow-up content, which is convenient for process management; the analysis of the process is convenient for formulating different sales strategies and providing reference for adjusting sales strategies; based on NLP technology, the sales and customer follow-up content, voice and text are classified to realize the conversion of unstructured content into structured classification. The structured classification can be calculated by data analysis tools to form charts and dashboard data, and the data can be made into a dashboard of the decision-making center, which is convenient and intuitive for sales and managers to make decisions on customer follow-up strategies. In addition, predictions are made through behavioral classification models, and the classification results of the sentiment analysis model are corrected, which can further improve the accuracy of decision-making.

[0073] Based on the above method embodiment, the present application embodiment also provides a device for processing customer follow-up information, see Figure 4 As shown, the device includes: an information acquisition module 402, which is used to acquire the customer follow-up information currently to be processed in the customer management system; a prediction module 404, which is used to perform sentiment analysis and prediction on the customer follow-up information through a preset sentiment analysis model, and output the sentiment classification label corresponding to the customer follow-up information; the sentiment classification label represents the degree of the customer's language intention; an indicator type determination module 406, which is used to determine the target customer indicator type corresponding to the customer follow-up information based on the sentiment classification labels corresponding to the customer follow-up information; a data update module 408, which is used to count the customer follow-up information into the existing dashboard data based on the sentiment classification labels and target customer indicator types corresponding to the customer follow-up information; the dashboard data includes statistical data under a single label dimension and / or a combination of multiple label dimensions, and a decision reminder module 410, which is used to make decision reminders to sales personnel after monitoring events that meet decision trigger conditions in the dashboard data.

[0074] Furthermore, the above-mentioned customer follow-up information includes: text communication information between the salesperson and the customer or text information obtained by converting voice communication information; the above-mentioned device also includes: a behavior analysis module, which is used to obtain customer behavior data corresponding to the customer follow-up information before the step of determining the target customer indicator type corresponding to the customer follow-up information based on the sentiment classification labels corresponding to the customer follow-up information; predict the customer behavior data through a preset behavior analysis model to obtain a behavior classification label corresponding to the customer behavior data; the behavior classification label represents the user's behavioral intention degree; judge whether the behavior classification label and the sentiment classification label are consistent; if they are inconsistent, correct the sentiment classification label with the behavior classification label; wherein the types corresponding to the sentiment classification label and the behavior classification label include: distrust, general, perfunctory, interested, high intention, disgust, competitor's iron fan, considering cooperation or already cooperating.

[0075] Furthermore, the above-mentioned device also includes: a model training module, which is used to execute the following sentiment analysis model training process: obtaining a business sample data set; the business sample data in the business sample data set includes: customer follow-up information and corresponding sentiment classification labels; applying the business sample data in the business sample data set to iteratively train the preset classification model based on the prediction results to obtain a sentiment analysis model.

[0076] Furthermore, the above-mentioned model training module is used to take each business sample data as the current sample data and perform the following prediction steps: output the prediction result corresponding to the current sample data through the preset classification model; if the prediction result corresponding to the current sample data does not meet expectations, perform data expansion and optimization on the current sample data to obtain new sample data; use the new sample data as the current sample data again, and continue to execute the prediction steps until the prediction accuracy of the preset classification model reaches the preset value; obtain the target sample data that can no longer be optimized, correct the prediction result of the target sample data based on the preset correction rules, and continue to iteratively train the preset classification model until the prediction accuracy of the model meets the business needs, and obtain the sentiment analysis model.

[0077] Furthermore, the above-mentioned model training module is used to match the prediction result corresponding to the target sample data with multiple preset rules; if the match is successful, the prediction value corresponding to the hit rule is used as the corrected prediction result corresponding to the target sample data.

[0078] Furthermore, the above-mentioned model training module is used to perform similarity matching on the prediction results of the first target sample data through a semantic similarity model for the first target sample data whose number of model prediction errors exceeds a threshold; a plurality of semantically similar sentences and corresponding prediction values ​​are defined in the semantic similarity model; and for the second target sample data whose model prediction errors are of a fixed type, a regular expression matching method is used to perform rule matching on the prediction results of the second target sample data.

[0079] Furthermore, a plurality of customer indicator types are pre-configured in the above-mentioned dashboard data; each customer indicator type corresponds to a target label dimension, or a target dimension combination containing multiple label dimensions; the indicator type determination module 406 is used to search for the type to which the sentiment classification label belongs from a plurality of customer indicator types according to the correspondence between the customer indicator type and the label; and the found type is determined as the target customer indicator type corresponding to the customer follow-up information.

[0080] Furthermore, the above-mentioned decision reminder module 410 is used to monitor events that meet the decision trigger conditions in the following ways: determine whether there is data exceeding the warning value in the statistical data under a single label dimension and / or the changing trend of the statistical data under the customer indicator type; determine whether the type and number of sentiment classification labels of a single customer within a specified time period meet the preset trigger conditions; generate corresponding decision reminder information according to the event type, and send it to the salesperson's terminal.

[0081] The device provided in the embodiment of the present application has the same implementation principle and technical effects as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the embodiment of the device, reference may be made to the corresponding contents in the aforementioned method embodiment.

[0082] The present application also provides an electronic device, such as Figure 5 As shown, it is a schematic diagram of the structure of the electronic device, wherein the electronic device includes a processor 51 and a memory 50, the memory 50 stores computer executable instructions that can be executed by the processor 51, and the processor 51 executes the computer executable instructions to implement the above method.

[0083] exist Figure 5 In the illustrated embodiment, the electronic device further includes a bus 52 and a communication interface 53 , wherein the processor 51 , the communication interface 53 and the memory 50 are connected via the bus 52 .

[0084] Among them, the memory 50 may include a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 53 (which can be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. can be used. The bus 52 can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 52 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0085] The processor 51 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 51. The above processor 51 can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to execute, or the hardware and software modules in the decoding processor are combined and executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor 51 reads the information in the memory and completes the steps of the method of the above embodiment in combination with its hardware.

[0086] An embodiment of the present application also provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the above method. The specific implementation can be found in the aforementioned method embodiment, which will not be repeated here.

[0087] The computer program products of the methods, devices, and electronic devices provided in the embodiments of the present application include a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the previous method embodiments. The specific implementation can be found in the method embodiments, which will not be repeated here.

[0088] Unless otherwise specifically stated, the relative steps, numerical expressions and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present application.

[0089] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0090] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0091] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for processing customer follow-up information, characterized in that: The method comprises: Obtain the customer follow-up information currently pending in the customer management system; Performing sentiment analysis and prediction on the customer follow-up information through a preset sentiment analysis model, and outputting a sentiment classification label corresponding to the customer follow-up information; the sentiment classification label represents the degree of the customer's language intention; Determining the target customer indicator type corresponding to the customer follow-up information based on the sentiment classification labels respectively corresponding to the customer follow-up information; Based on the sentiment classification labels and the target customer indicator types respectively corresponding to the customer follow-up information, the customer follow-up information is counted into the existing dashboard data; the dashboard data includes statistical data under a single label dimension and / or a combination of multiple label dimensions, After an event that meets the decision triggering condition is detected in the dashboard data, a decision reminder is given to the salesperson.

2. The method according to claim 1, characterized in that The customer follow-up information includes: text communication information between the salesperson and the customer or text information converted from voice communication information; before the step of determining the target customer indicator type corresponding to the customer follow-up information based on the sentiment classification labels respectively corresponding to the customer follow-up information, the method further includes: If the customer behavior data corresponding to the customer follow-up information is obtained, the customer behavior data is predicted by a preset behavior analysis model to obtain a behavior classification label corresponding to the customer behavior data; the behavior classification label represents the degree of the user's behavior intention; Determining whether the behavior classification label and the emotion classification label are consistent; If they are inconsistent, the emotion classification label is corrected with the behavior classification label; Among them, the types corresponding to the sentiment classification labels and the behavior classification labels include: distrust, general, perfunctory, interested, high intention, disgust, competitor fan, considering cooperation or already cooperating.

3. The method according to claim 1, characterized in that The training process of the sentiment analysis model is as follows: Acquire a business sample data set; the business sample data in the business sample data set includes: customer follow-up information and corresponding sentiment classification labels; The business sample data in the business sample data set is applied to perform iterative training on the preset classification model for correction based on the prediction results to obtain the sentiment analysis model.

4. The method according to claim 3, characterized in that The step of applying the business sample data in the business sample data set to iteratively train the preset classification model based on the prediction results to obtain the sentiment analysis model includes: Take each business sample data as the current sample data and perform the following prediction steps: Output the prediction result corresponding to the current sample data through the preset classification model; If the prediction result corresponding to the current sample data does not meet expectations, perform data expansion and optimization on the current sample data to obtain new sample data; use the new sample data as the current sample data again, and continue to perform the prediction step until the prediction accuracy of the preset classification model reaches a preset value; The target sample data that cannot be further optimized is obtained, the prediction result of the target sample data is corrected based on the preset correction rule, and the preset classification model is continuously iterated to be trained until the prediction accuracy of the model meets the business requirements, thereby obtaining the sentiment analysis model.

5. The method according to claim 4, characterized in that The step of correcting the prediction result of the target sample data based on a preset correction rule includes: Matching the prediction result corresponding to the target sample data with a plurality of preset rules; If the match is successful, the prediction value corresponding to the hit rule is used as the corrected prediction result corresponding to the target sample data.

6. The method according to claim 5, characterized in that The step of matching the prediction result corresponding to the target sample data with a plurality of preset rules comprises: For first target sample data whose number of model prediction errors exceeds a threshold, similarity matching is performed on the prediction results of the first target sample data through a semantic similarity model; the semantic similarity model defines a plurality of semantically similar sentences and corresponding prediction values; For the second target sample data whose model prediction error is of a fixed type, a regular expression matching method is used to perform rule matching on the prediction result of the second target sample data.

7. The method according to claim 1, characterized in that There are multiple customer indicator types pre-configured in the dashboard data; each customer indicator type corresponds to a target label dimension, or a target dimension combination containing multiple label dimensions; the target label dimensions include: The step of determining the target customer indicator type corresponding to the customer follow-up information based on the sentiment classification labels respectively corresponding to the customer follow-up information includes: According to the correspondence between the customer indicator type and the label, searching for the type to which the sentiment classification label belongs from a plurality of customer indicator types; The found type is determined as the target customer indicator type corresponding to the customer follow-up information.

8. The method according to claim 1, characterized in that After an event that meets the decision triggering condition is detected in the dashboard data, the step of reminding the salesperson to make a decision includes: Events that meet decision trigger conditions are monitored in the following ways: Determine whether there is data exceeding the warning value in the statistical data under a single tag dimension and / or the change trend of the statistical data under the customer indicator type; Determine whether the type and number of sentiment classification labels of a single customer within a specified time period meet the preset trigger conditions; According to the event type, corresponding decision reminder information is generated and sent to the salesperson's terminal.

9. A device for processing customer follow-up information, characterized in that: The device comprises: The information acquisition module is used to obtain the customer follow-up information currently to be processed in the customer management system; A prediction module, used to perform sentiment analysis and prediction on the customer follow-up information through a preset sentiment analysis model, and output a sentiment classification label corresponding to the customer follow-up information; the sentiment classification label represents the degree of the customer's language intention; An indicator type determination module, used to determine the target customer indicator type corresponding to the customer follow-up information based on the sentiment classification labels respectively corresponding to the customer follow-up information; A data updating module is used to count the customer follow-up information into the existing dashboard data based on the sentiment classification labels and the target customer indicator types corresponding to the customer follow-up information; the dashboard data includes statistical data under a single label dimension and / or a combination of multiple label dimensions, The decision reminder module is used to remind the sales staff of the decision after detecting an event that meets the decision triggering condition in the dashboard data.

10. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the method according to any one of claims 1 to 8.

Images