Intelligent recommendation method for welcome language and first-time contact verbal skill, medium and electronic equipment

Abstract

The invention provides an intelligent recommendation method for welcome words and first-time contact verbal skills, a medium and electronic equipment. The method comprises the following steps: acquiring historical chat data, wherein the historical chat data comprises historical welcome language data and historical first contact verbal skill data of different channel incoming lines; in response to the access to the chat channel, obtaining a high-score welcome based on the historical welcome data, performing language structure analysis processing on the high-score welcome, obtaining an expression pattern of the high-score welcome, and generating a welcome corresponding to the current channel based on the expression pattern of the high-score welcome; and when reply chat data of the user to the welcome language is received, obtaining a category and verbal skill content corresponding to the high-score first-time contact verbal skill based on historical first-time contact verbal skill data, and recommending the category and verbal skill content corresponding to the high-score first-time contact verbal skill to the user, so that the user edits the first-time contact verbal skill corresponding to the current channel. According to the method and the device, efficient, accurate and intelligent recommendation of welcome words and first-time contact verbal skills can be realized.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, and in particular to the field of intelligent information recommendation technology in the field of smart investment consulting. Background Art

[0002] In the investment advisory sector, when employees first connect with clients, sending a welcome message and initial contact scripts via chat tools is a critical customer service step. The quality of this initial interaction directly impacts client trust in the service and the potential for subsequent investment services. Because customer needs vary significantly across channels and chat tools, precisely designed welcome messages and initial contact scripts tailored to channel characteristics and client profiles are essential to effectively drive client conversions. However, in industrial practice, the evaluation of welcome messages and initial contact scripts faces the challenges of diverse fields and tasks. Chat scenarios and data characteristics vary across different fields, making it difficult for a general approach to adapt to the specific needs of the investment advisory sector.

[0003] In the prior art, there are mainly two methods, but both have obvious limitations:

[0004] The single evaluation method based on customer responses judges the effectiveness of the welcome message by the content of the customer's response. For example, customers who respond positively are considered to have a successful welcome message. However, this method is relatively limited and ignores more information that may affect the quality of customer interaction, such as the language composition of the welcome message itself, the time interval between customer responses (such as quick responses may indicate high interest), and whether the customer actually purchases investment advisory services later (key indicators). The one-sidedness of this method makes it impossible to fully evaluate the effectiveness of the welcome message, thereby limiting the room for optimization. Another method is a cluster matching method based on basic customer information. Customers are clustered according to their basic information (such as WeChat source, nickname, wealth value, age group, etc.), and then matched with different welcome messages. This method is more comprehensive than a single evaluation, but there are still problems. It ignores the potential semantic information in the chat data and lacks in-depth analysis of the long-term effects of the welcome message (such as service conversion rate).

[0005] Patent CN 115630206 A, a prior art patent, proposes a welcome message matching method that adaptively matches welcome messages based on customer origin. However, this patent's technical solution ignores the impact of the welcome message's content on customer response behavior, making it difficult to comprehensively evaluate the effectiveness of the welcome message. It also fails to quantify the impact of different welcome messages on customer response rates, and thus fails to guide companies in optimizing welcome messages to increase customer engagement. Furthermore, this patent fails to analyze service conversion rates, failing to establish a link between welcome messages, customer interaction behavior, and ultimate service conversion rates, making it impossible to fully assess the actual contribution of welcome messages to business objectives.

[0006] Therefore, it is necessary to design a recommendation method that can adapt to the investment advisory service scenario to support efficient and accurate communication with customers. Summary of the Invention

[0007] The present application provides a method, medium, and electronic device for intelligently recommending welcome words and first-contact scripts for users, which are used to intelligently recommend welcome words and first-contact scripts for users efficiently and accurately.

[0008] To achieve the above-mentioned purpose and other related purposes, the first aspect of the present application provides an intelligent recommendation method for welcome words and first contact scripts, comprising: obtaining historical chat data, wherein the historical chat data includes historical welcome word data and historical first contact script data of different channels; in response to accessing the chat channel, obtaining the average reply time, reply rate and transaction rate of each welcome word based on the historical welcome word data, and obtaining a comprehensive score for each welcome word based on the average reply time, the reply rate and the transaction rate; obtaining high-scoring welcome words according to the high or low comprehensive score of each welcome word, performing language structure analysis on the high-scoring welcome words, obtaining the expression pattern of the high-scoring welcome words, and The expression pattern of the welcome words generates a welcome word corresponding to the current channel; when the user's reply chat data to the welcome words is received, the average reply time, reply rate and transaction rate of each type of first contact script are respectively obtained based on the historical first contact script data, and the comprehensive score of each type of first contact script is obtained based on the average reply time, the reply rate and the transaction rate of each type of first contact script; according to the high or low comprehensive score of each type of first contact script, the category and script content corresponding to the high-scoring first contact script are obtained, and the category and script content corresponding to the high-scoring first contact script are recommended to the user, so that the user can edit the first contact script corresponding to the current channel according to the category and script content corresponding to the high-scoring first contact script.

[0009] In some embodiments of the first aspect of the present application, according to the intelligent recommendation method for welcome messages and first contact scripts according to claim 1, the average reply time for each welcome message is obtained by:

[0010]

[0011] Among them, T i is the average reply time for each welcome message, T is a non-empty reply time set, min(T) is the shortest reply time, and max(T) is the longest reply time;

[0012] The response rate of each welcome message is obtained in the following ways:

[0013]

[0014] Among them, R i is the response rate of each welcome message, C i N_C is the total count of each greeting message. i The number of greeting messages whose reply content is not empty;

[0015] The methods for obtaining the conversion rate of each welcome message include:

[0016]

[0017] Among them, O i is the transaction rate of each welcome message, O_C i The number of service transactions for each welcome message is the number of service transaction times that are not empty.

[0018] In some embodiments of the first aspect of the present application, obtaining a comprehensive score for each welcome message includes: normalizing the average reply time, the reply rate, and the transaction rate of each welcome message to obtain the normalized average reply time, the normalized reply rate, and the normalized transaction rate; configuring corresponding weight coefficients for the normalized average reply time, the normalized reply rate, and the normalized transaction rate; wherein the sum of the weight coefficients is 1; the comprehensive score for each welcome message is the sum of the normalized average reply time, the normalized reply rate, and the normalized transaction rate multiplied by their respective weight coefficients.

[0019] In some embodiments of the first aspect of the present application, the language structure analysis and processing of the high-scoring welcome words to obtain the expression pattern of the high-scoring welcome words includes: accepting input welcome word analysis prompt words; calling the AI ​​big model, and the AI ​​big model performs language structure analysis and processing on the high-scoring welcome words according to the analysis requirements of the welcome word analysis prompt words to obtain the expression pattern of the high-scoring welcome words.

[0020] In some embodiments of the first aspect of the present application, obtaining the average response time, response rate and transaction rate of each type of first contact speech based on the historical first contact speech data includes: accepting input first contact speech analysis prompts; calling the AI ​​big model, the AI ​​big model classifies the historical first contact speech data according to the analysis requirements of the first contact speech analysis prompts, and obtains the category of the first contact speech; obtaining the average response time, response rate and transaction rate of each type of first contact speech.

[0021] In some embodiments of the first aspect of the present application, the method for obtaining the average response time for each type of first contact speech includes:

[0022]

[0023] Among them, T i is the average response time for each type of first contact script, T is a non-empty set of response times, min(T) is the shortest response time, and max(T) is the longest response time;

[0024] The response rate for each type of first contact script is obtained in the following ways:

[0025]

[0026] Among them, R i is the response rate of each type of first contact script, C i N_C is the total number of first contact scripts for each type. i The number of first contact scripts of each type whose response content is not empty;

[0027] The methods for obtaining the closing rate of each type of first contact script include:

[0028]

[0029] Among them, O i is the closing rate of each type of first contact script, O_C i The number of service transactions for each type of first contact script is the number of service transaction times that are not empty.

[0030] In some embodiments of the first aspect of the present application, obtaining the comprehensive score of each type of first contact script includes: normalizing the average response time, the response rate, and the transaction rate of each type of first contact script to obtain the normalized average response time, the normalized response rate, and the normalized transaction rate; configuring corresponding weight coefficients for the normalized average response time, the normalized response rate, and the normalized transaction rate; wherein the sum of the weight coefficients is 1; the comprehensive score of each type of first contact script is the sum of the normalized average response time, the normalized response rate, and the normalized transaction rate multiplied by their respective weight coefficients.

[0031] To achieve the above-mentioned purpose and other related purposes, the second aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, and the computer program implements the method when executed by a processor.

[0032] To achieve the above-mentioned purpose and other related purposes, the third aspect of the present application provides a computer program product, which includes computer program code. When the computer program code is run on a computer, the computer implements the method.

[0033] To achieve the above-mentioned purpose and other related purposes, the fourth aspect of the present application provides an electronic terminal, comprising a memory, a processor, and a computer program stored in the memory; the processor executes the computer program to implement the method.

[0034] The intelligent recommendation method for welcome words and first contact dialogue provided by the embodiment of the present application has the following beneficial effects:

[0035] This application implements efficient and accurate intelligent recommendations for welcome messages and first-contact scripts, improving the overall service level and effectively avoiding the problem that traditional script optimization relies on subjective experience and is difficult to quantify. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 Shown is a flowchart of a method for intelligently recommending welcome messages and first-time contact scripts according to an embodiment of the present application.

[0037] Figure 2 Shown is a flowchart of obtaining a comprehensive score for each welcome message in an intelligent recommendation method for welcome messages and first contact scripts according to an embodiment of the present application.

[0038] Figure 3 Shown is a schematic diagram of the principle of obtaining the welcome message expression pattern in the intelligent recommendation method for welcome messages and first contact dialogue according to an embodiment of the present application.

[0039] Figure 4 Shown is a flowchart of obtaining parameters of the first contact script in the intelligent recommendation method of the welcome message and the first contact script according to one embodiment of the present application.

[0040] Figure 5 Shown is a schematic diagram of the principle of obtaining the category of the first contact script in the intelligent recommendation method of the welcome message and the first contact script according to one embodiment of the present application.

[0041] Figure 6 Shown is a flowchart of obtaining a comprehensive score for first contact scripts in an intelligent recommendation method for welcome messages and first contact scripts according to an embodiment of the present application.

[0042] Figure 7 Shown is a schematic diagram of the overall implementation process of the intelligent recommendation method for welcome words and first contact dialogue according to one embodiment of the present application.

[0043] Figure 8 Shown is a structural schematic diagram of an electronic device in one embodiment of the present application.

[0044] Component number description

[0045] 100 electronic devices

[0046] 101 Memory

[0047] 102 processors

[0048] 103 Display

[0049] Steps S100 to S500

[0050] Steps S210 to S230

[0051] Steps S410 to S430

[0052] Steps S441 to S443 DETAILED DESCRIPTION

[0053] The following describes the embodiments of the present application through specific examples. Those skilled in the art can easily understand the other advantages and effects of the present application from the content disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0054] Before further explaining the present invention in detail, the nouns and terms involved in the embodiments of the present invention are explained. The nouns and terms involved in the embodiments of the present invention are subject to the following interpretations:

[0055] <1> Welcome message is the first sentence sent to customers after they access the chat channel (such as WeChat account), usually sent to a group of customers.

[0056] <2> The first contact script is the first sentence sent by the user after the customer responds to the welcome message, usually in a one-on-one chat.

[0057] This embodiment provides an intelligent recommendation method, medium and electronic device for welcome words and first contact scripts, which are applied to communication and chat scenarios. This embodiment analyzes the effects of welcome words and contact scripts on different channels. The system automatically identifies highly rated language structures and expression patterns, and guides employees to send more attractive and conversion-capable scripts, thereby improving customer response rates and service transaction rates, and realizing refined operations.

[0058] This embodiment provides an intelligent recommendation method, medium, and electronic device for welcome messages and first-time contact scripts. First, the method obtains and structures chat data between company employees and customers over the past month, focusing on interactions within 24 hours after a customer calls. Next, for welcome messages across different channels, the method collects statistics on welcome message frequency, customer response rate, average response time, and transaction rate. This data is normalized and weighted averaged to obtain a comprehensive score and ranking, identifying effective welcome messages. A large language model is then used to analyze the structure of high-scoring scripts and extract expression patterns. These high-scoring expression patterns and scripts are then used for business use. The large model is then used to classify the language content of first-time contact scripts and identify script types (e.g., download guidance, account opening guidance, etc.). The method analyzes the effectiveness indicators of each script, including response rate, average response time, and transaction rate, normalizes these metrics, and calculates a weighted average sum as the final score. Finally, the method outputs high-scoring contact script categories and related script content for business personnel to use based on channel requirements, improving service efficiency and conversion rates. Through the combination of the above technical features, this embodiment effectively avoids the problem that traditional speech optimization relies on subjective experience and is difficult to quantify, and guides business personnel to send more attractive and conversion-capable speech, thereby improving customer response rate and service transaction rate, and realizing refined operations.

[0059] The following is a combination of the appended examples of the present application Figure 1 To the attached Figure 8 , the technical solutions in the embodiments of the present application are described in detail, so that those skilled in the art can understand and implement the intelligent recommendation method for welcome words and first contact words in this embodiment without creative work.

[0060] Figure 1 Shown is a flow chart of the intelligent recommendation method for welcome words and first contact words in the embodiment of this application. Figure 1 As shown, the intelligent recommendation method for welcome words and first contact words provided in the embodiment of the present application includes the following steps S100 to S500.

[0061] Step S100: Acquire historical chat data, including historical welcome message data and historical first contact dialogue data for incoming calls through different channels;

[0062] Step S200, in response to accessing the chat channel, obtaining the average reply time, reply rate, and transaction rate of each welcome message based on the historical welcome message data, and obtaining a comprehensive score for each welcome message based on the average reply time, the reply rate, and the transaction rate;

[0063] Step S300: obtaining high-scoring welcome messages based on the comprehensive scores of each of the welcome messages, performing language structure analysis on the high-scoring welcome messages to obtain expression patterns of the high-scoring welcome messages, and generating a welcome message corresponding to the current channel based on the expression patterns of the high-scoring welcome messages;

[0064] Step S400, when receiving the chat data of the user's reply to the welcome message, respectively obtain the average reply time, reply rate and transaction rate of each type of first contact script based on the historical first contact script data, and obtain the comprehensive score of each type of first contact script based on the average reply time, the reply rate and the transaction rate of each type of first contact script; Step S500, according to the high or low comprehensive score of each type of first contact script, obtain the category and script content corresponding to the high-scoring first contact script, and recommend the category and script content corresponding to the high-scoring first contact script to the user, so that the user can edit the first contact script corresponding to the current channel according to the category and script content corresponding to the high-scoring first contact script.

[0065] The intelligent recommendation method for welcome words and first-contact scripts in this embodiment assists users in accurately selecting or adjusting script content according to customers from different channels. From customer access, reply, transaction to script optimization, a complete closed loop from data collection, analysis, evaluation to execution is constructed, which enables efficient and accurate intelligent recommendation of welcome words and first-contact scripts, improves communication efficiency and customer experience, provides reliable support for intelligent services in the investment advisory industry, and effectively avoids the problem that traditional script optimization relies on subjective experience and is difficult to quantify.

[0066] The following describes in detail the above steps S100 to S500 of the intelligent recommendation method for welcome words and first contact dialogue in this embodiment.

[0067] Step S100: Acquire historical chat data, wherein the historical chat data includes historical welcome data and historical first contact speech data of different channels.

[0068] The historical chat data is preferably chat data between users and customers over a recent period of time. For example, historical chat data from customers who have connected with the company within the past month is selected. In other words, in this embodiment, chat data between company employees and customers over the past month is obtained and structured, focusing on the interactions within 24 hours after the customer connects with the company.

[0069] In this embodiment, the historical chat data is also processed into structured data. For example, with respect to the contact situation within 24 hours after the call is made, the historical welcome data and the historical first contact speech data of the calls made through different channels will be organized into structured data. The structured data includes fields including but not limited to: employee name, customer name, call time, service transaction time (can be empty), channel, welcome, customer's reply to the welcome (empty if the customer does not reply), time interval for the customer to reply to the welcome (empty if the customer does not reply, in seconds), first contact speech, customer's reply to the first contact (empty if the customer does not reply), time interval for the customer to reply to the first contact (empty if the customer does not reply, in seconds), etc.

[0070] Step S200, in response to accessing the chat channel, obtain the average reply time, reply rate and transaction rate of each welcome message based on the historical welcome message data, and obtain a comprehensive score for each welcome message based on the average reply time, the reply rate and the transaction rate.

[0071] In this embodiment, after obtaining the welcome message data of incoming calls from different channels, for the welcome messages of different channels, in this embodiment, statistics are collected on the welcome message frequency, customer response rate, average response time, and transaction rate data, and finally a comprehensive score is obtained for each welcome message.

[0072] In a specific implementation of this embodiment, the method for intelligently recommending welcome messages and first contact scripts includes obtaining the average reply time for each welcome message by:

[0073]

[0074] Among them, T i is the average reply time for each welcome message, T is a non-empty reply time set, min(T) is the shortest reply time, and max(T) is the longest reply time.

[0075] In this embodiment, the fields of welcome word count and average welcome word reply time are added. The welcome word count is a cluster count of repeated welcome words, and the count of each welcome word is C i The average response time field for the welcome message is T i The calculation method is to remove the data with empty values ​​of customer response time for the welcome message and the shortest time and the longest time, and then take the average response time (take minutes as the unit, take 1 minute if it is within 1 minute, and round up if it exceeds 1 minute).

[0076] Clustering and counting repeated welcome phrases involves converting the welcome phrase text into a vector representation using TF-IDF (Term Frequency-Inverse Document Frequency), Word2Vec, or BERT to facilitate similarity calculation. For example, the welcome phrase text is converted into a sparse matrix, or the word vectors of each sentence are averaged or pooled to obtain a sentence-level vector. The similarity of the welcome phrase text is calculated using methods such as cosine similarity or Jaccard similarity. Finally, a similarity threshold is set and a clustering algorithm (K-Means, DBSCAN, or hierarchical clustering) is used to cluster and count duplicate welcome phrases.

[0077] For each welcome message, let its non-empty reply time set be:

[0078] T={t1,t2,...,t n}(length is N), shortest time: min(T), longest time: max(T), average response time is calculated as:

[0079]

[0080] In this embodiment, a new response rate field is added to calculate the response rate R of each greeting message. i The number of customer responses that are not empty is N_C i The response rate of each welcome message is obtained in the following ways:

[0081]

[0082] Among them, R i is the response rate of each welcome message, C i N_C is the total count of each greeting message. i The number of greeting messages whose reply content is not empty.

[0083] In this embodiment, a new transaction rate field is added to calculate the transaction rate of each welcome message. i The number of service transactions for each welcome message is the number of service transaction times that are not empty, assuming it is O_C i The methods for obtaining the conversion rate of each welcome message include:

[0084]

[0085] Among them, O i is the transaction rate of each welcome message, O_C i The number of service transactions for each welcome message is the number of service transaction times that are not empty.

[0086] In this embodiment, a new welcome message comprehensive score field is added to calculate the comprehensive score S of each welcome message. i . Figure 2Shown is a flow chart of obtaining a comprehensive score for each welcome message in the intelligent recommendation method for welcome messages and first contact speech in one embodiment of the present application. Figure 2 As shown, in a specific implementation of this embodiment, obtaining the comprehensive score of each welcome message includes the following steps S210 to S230.

[0087] Step S210: normalize the average reply time, the reply rate, and the transaction rate of each greeting message to obtain the normalized average reply time, the normalized reply rate, and the normalized transaction rate;

[0088] Step S220 , configuring corresponding weight coefficients for the normalized average reply time, the normalized reply rate, and the normalized transaction rate, respectively; wherein the sum of the weight coefficients is 1;

[0089] In step S230 , the comprehensive score of each welcome message is the sum of the normalized average reply time, the normalized reply rate, and the normalized transaction rate multiplied by their respective weight coefficients.

[0090] In this embodiment, the average customer response time, response rate, and transaction rate of each welcome message are normalized. The normalized average response time T norm The calculation method is:

[0091]

[0092] Among them, T max The longest average time for customers to respond to the welcome message, T min The shortest average time for customers to respond to the welcome message, T i The average time it takes for customers to reply to this welcome message.

[0093] In this embodiment, the normalized recovery rate R norm The calculation method is:

[0094]

[0095] Among them, R max is the highest response rate of the welcome message, R min is the minimum response rate of the welcome message, R i is the response rate of this welcome message.

[0096] In this embodiment, the normalized transaction rate O norm The calculation method is:

[0097]

[0098] Among them, O max The highest transaction rate for the welcome message, Omin is the lowest transaction rate of the welcome message, O i The conversion rate of this welcome message.

[0099] The overall score of each welcome message is S i Calculation method:

[0100] S i =W T *T norm +W R *R norm +W O *O norm

[0101] Where: W T +W R +W O =1,W T , W R , W O It can be automatically adjusted according to actual business, for example, R =0.5,W O =0.4,W T =0.1, the business focuses on the response rate and can R Increase the service transaction rate and focus on W O Adjust it higher, focus on the reply experience and you can T In this embodiment, by flexibly setting scoring weight parameters, such as focusing on reply rate, transaction rate, or reply experience, the scoring system can be automatically optimized according to actual business scenarios to meet the needs of speech optimization under different operation strategies.

[0102] Step S300: obtaining high-scoring welcome messages according to the comprehensive scores of each of the welcome messages, performing language structure analysis on the high-scoring welcome messages to obtain expression patterns of the high-scoring welcome messages, and generating a welcome message corresponding to the current channel based on the expression patterns of the high-scoring welcome messages.

[0103] Performing language structure analysis on the high-scoring welcome message to obtain the expression pattern of the high-scoring welcome message includes:

[0104] 1) Perform syntactic analysis on high-scoring welcome messages: Use natural language processing tools (such as SpaCy, NLTK, or BERT) to perform syntactic analysis on the welcome messages and extract the subject, predicate, object, modifiers, and other components of the sentence.

[0105] 2) Pattern recognition of high-scoring welcome messages: By analyzing the commonalities of high-scoring welcome messages, we can extract their expression patterns, such as greetings, emotional expressions, and channel-related vocabulary.

[0106] 3) Pattern extraction of high-scoring welcome messages: Use machine learning algorithms (such as hidden Markov models and conditional random fields) to automatically extract expression patterns from high-scoring welcome messages.

[0107] 4) Generate a pattern expression template based on the extracted pattern.

[0108] Then, based on the current channel and target audience, fill in the variables in the edit mode expression template to generate a welcome message corresponding to the channel.

[0109] Figure 3 Shown is a schematic diagram of the principle of obtaining the welcome expression pattern in the intelligent recommendation method of the welcome words and first contact words in one embodiment of the present application. Figure 3 As shown, in a specific implementation method of this embodiment, the language structure analysis processing of the high-scoring welcome words to obtain the expression pattern of the high-scoring welcome words includes: accepting input welcome word analysis prompt words; calling the AI ​​big model, and the AI ​​big model performs language structure analysis processing on the high-scoring welcome words according to the analysis requirements of the welcome word analysis prompt words to obtain the expression pattern of the high-scoring welcome words.

[0110] In this embodiment, according to the comprehensive score of each welcome message obtained, the scores are ranked from high to low, and welcome messages with high comprehensive scores can be obtained. A language structure analysis is performed on one or more welcome messages with high rankings. The number of welcome messages is based on the actual number. For example, a language structure analysis is performed on the top 10 welcome messages, and prompt words are written such as: You are a professional key point extraction expert who can extract key points based on the speech skills and require conciseness. The specific prompt words can be adjusted according to actual conditions and the AI ​​large model can be called. Including but not limited to models such as deepseek-R1, Qwen-max, Qwen2.5-32B, the expression pattern of the welcome message can be obtained. For example, the expression pattern of a high-scoring welcome message:

[0111] (1) Polite words, (2) Introduce the company, a long-established investment institution, (3) Introduce the employee (employee number, practice number), (4) Introduce the service (information, etc.), (5) Explain that you are not a robot.

[0112] When users send welcome messages according to different channels, they can refer to the above-mentioned high-rated welcome message expression models and content, modify them and send them to customers.

[0113] Step S400, when receiving the chat data of the user's reply to the welcome message, the average reply time, reply rate and transaction rate of each type of first contact script are obtained based on the historical first contact script data, and the comprehensive score of each type of first contact script is obtained based on the average reply time, reply rate and transaction rate of each type of first contact script.

[0114] Figure 4 Shown is a flow chart of obtaining parameters of the first contact speech in the intelligent recommendation method of the welcome words and first contact speech in one embodiment of the present application. Figure 4 As shown, in a specific implementation of this embodiment, the step of obtaining the average reply time, reply rate, and transaction rate of each type of first contact speech based on the historical first contact speech data includes the following steps S410 to S430.

[0115] Step S410: accepting the input first contact analysis prompt words;

[0116] Step S420: calling the AI ​​big model, wherein the AI ​​big model classifies the historical first contact speech data according to the analysis requirements of the first contact speech analysis prompt words to obtain the category of the first contact speech;

[0117] Step S430 , respectively obtain the average response time, response rate, and transaction rate of each type of first contact script.

[0118] This embodiment uses large language models (such as Qwen, DeepSeek, etc.) to perform language structure extraction and speech classification, which can extract core expression elements and potential categories from unstructured employee language, thereby promoting intelligent classification and continuous optimization of contact speech.

[0119] Figure 5 The diagram shows the principle of obtaining the category of the first contact words in the intelligent recommendation method of the welcome words and first contact words in one embodiment of the present application. Figure 5 As shown, the user-entered first contact speech analysis prompt is input into the AI ​​big model. The AI ​​big model classifies the historical first contact speech data based on the analysis requirements of the first contact speech analysis prompt, and then outputs the first contact speech category. In this embodiment, the first contact speech category field can be a single category or a combination of multiple categories.

[0120] In this embodiment, a new first contact speech category field is added to process the first contact speech, analyze the language structure of the contact speech, and classify the speech by writing prompt words using AI large models, including but not limited to qwen-plus and qwen-turbo models. The input first contact speech analysis prompt words are as follows:

[0121] You are a professional classification expert, classifying texts. The existing categories are: greeting, guidance for downloading software,

[0122] It may be necessary to explore more categories, extract them from the customer's actual problems, and be brief without unnecessary explanations.

[0123] Among them, the specific first contact speech analysis prompt words can be adjusted according to the actual speech content.

[0124] The AI ​​model classifies the historical first contact conversation data based on the analysis requirements of the first contact conversation analysis prompt words to obtain the categories of the first contact conversation. For example, all categories are obtained, such as: guidance on downloading and using software (indicator mini-program), free stocks, guidance on stock diagnosis, guidance on cooperation, inquiry on account opening, information-related, etc. The first contact conversation may be a single category or a combination of multiple categories, such as: information-related / guidance on downloading and using software, indicator mini-program.

[0125] In a specific implementation of this embodiment, the method for obtaining the average response time for each type of first contact speech includes:

[0126]

[0127] Among them, T i is the average response time for each type of first contact script, T is a non-empty set of response times, min(T) is the shortest response time, and max(T) is the longest response time

[0128] In this embodiment, clustering is performed based on the first contact speech category. New fields are added: category count and average response time of contact speech. The count of each type of speech is C i The average response time field for contact dialogue is T i The calculation method is to remove data with empty response times for first contact, and then take the average response time after removing the shortest and longest response times. For example, if the unit is minutes, 1 minute is taken as the response time, and any time exceeding 1 minute is rounded up.

[0129] For each type of first contact script, let the set of non-empty response times be:

[0130] T={t1,t2,...,t n}(length is N), shortest time: min(T), longest time: max(T), average response time for first contact is calculated as:

[0131]

[0132] In this embodiment, a new response rate field is added to calculate the response rate R of each type of contact words. i The number of customers whose first contact reply is not empty is N_C i The response rate of each type of first contact script is obtained in the following ways:

[0133]

[0134] Among them, Ri is the response rate of each type of first contact script, C i N_C is the total number of first contact scripts for each type. i The number of first contact scripts of each type whose response content is not empty;

[0135] In this embodiment, a new transaction rate field is added to calculate the transaction rate of each type of first contact speech. i The number of service transactions for each type of first contact script is the number of service transaction times that are not empty, assuming it is O_C i The methods for obtaining the closing rate of each type of first contact script include:

[0136]

[0137] Among them, O i is the closing rate of each type of first contact script, O_C i The number of service transactions for each type of first contact script is the number of service transaction times that are not empty.

[0138] In this embodiment, a new field for comprehensive score of first contact words is added to calculate the comprehensive score S of each type of contact words. i , Figure 6 Shown is a flow chart of obtaining a comprehensive score of the first contact speech in the intelligent recommendation method of the welcome words and first contact speech in one embodiment of the present application. Figure 6 As shown, in a specific implementation of this embodiment, obtaining the comprehensive score of each type of first contact speech includes the following steps S441 to S443.

[0139] Step S441 , normalizing the average response time, response rate, and transaction rate for each type of first contact script to obtain a normalized average response time, normalized response rate, and normalized transaction rate;

[0140] Step S442 , respectively configuring corresponding weight coefficients for the normalized average reply time, the normalized reply rate, and the normalized transaction rate; wherein the sum of the weight coefficients is 1;

[0141] In step S443 , the comprehensive score of each type of first contact script is the sum of the normalized average response time, the normalized response rate, and the normalized transaction rate multiplied by their respective weight coefficients.

[0142] In this embodiment, the average customer response time, response rate, and transaction rate of each type of first contact script are normalized. The normalized average response time T for each type of first contact script is norm The calculation method is:

[0143]

[0144] Among them, T max The longest average time to respond to first-time customers, T min The shortest average response time for first-time contact customers, T i The average response time for first-time customers using this type of customer service.

[0145] In this embodiment, the normalized response rate of the first contact technique R norm The calculation method is:

[0146]

[0147] Among them, R max The highest response rate for the first contact script is R min is the minimum response rate of the first contact script, R i The response rate for this type of first contact script.

[0148] In this embodiment, the normalized closing rate of the first contact technique is O norm The calculation method is:

[0149]

[0150] Among them, O max The highest closing rate for first contact words, O min The lowest closing rate for the first contact, O i The conversion rate of this type of first-time contact script.

[0151] In this embodiment, the comprehensive score S of each type of first contact speech is i The calculation method is:

[0152] S i =W T *T norm +W R *R norm +W O *O norm

[0153] W T +W R +W O =1

[0154] Among them, W T , W R , W O It can be automatically adjusted according to actual business. This solution takes W R =0.5,W O =0.4,W T =0.1, the business focuses on the response rate and can RIncrease the service transaction rate and focus on W O Adjust it higher, focus on the reply experience and you can T In this embodiment, by flexibly setting scoring weight parameters, such as focusing on reply rate, transaction rate, or reply experience, the scoring system can be automatically optimized according to actual business scenarios to meet the needs of speech optimization under different operation strategies.

[0155] Step S500, obtaining the categories and speech contents corresponding to the high-scoring first contact speech according to the comprehensive scores of each type of first contact speech, and recommending the categories and speech contents corresponding to the high-scoring first contact speech to the user, so that the user can edit the first contact speech corresponding to the current channel according to the categories and speech contents corresponding to the high-scoring first contact speech.

[0156] Based on the overall scores of the first contact scripts, the scores are ranked from high to low to identify the first contact script categories with the highest overall scores. These high-scoring first contact script categories and related script content are provided as a reference for business system users. After the customer responds to the welcome message, users can edit or modify these high-scoring first contact script categories and content based on different channels and send them to the customer.

[0157] Figure 7 Shown is a schematic diagram of the overall implementation process of the intelligent recommendation method for the welcome message and the first contact speech in one embodiment of the present application. Figure 7 As shown, the implementation process of the intelligent recommendation method for the welcome message and the first contact dialogue of this embodiment is as follows:

[0158] Obtain historical chat data: Obtain and structure the chat data between company employees and customers for the past month, focusing on the interaction content within 24 hours after the customer calls.

[0159] The historical welcome message data and first contact script data for calls from different channels will be organized into structured data, including but not limited to: employee name, customer name, call time, service transaction time, channel, welcome message, customer response to the welcome message, time interval between customer responses to the welcome message, first contact script, customer response to the first contact (and time interval between customer responses to the first contact), etc.

[0160] We conduct statistical analysis of welcome messages across different channels, including frequency, customer response rate, average response time, and transaction rate. We then normalize the average response time, response rate, and transaction rate for each welcome message, calculate a comprehensive score, and rank the welcome messages. This normalization and weighted average of the above data yields a comprehensive score and ranking, allowing us to identify effective welcome messages.

[0161] The large language model is then used to extract the key points of the highly rated speech, and output the expression pattern and content of the highly rated welcome message for user reference and editing.

[0162] After sending a welcome message to the customer, if you receive a reply from the customer, continue to recommend the first contact script to the user.

[0163] Statistical analysis is performed on first-contact conversation data. Prompt phrases used in first-contact conversation analysis are fed into a large AI model. This model then classifies language content, identifies conversation types, and generates category fields, such as download guidance and account opening guidance. Effectiveness metrics for each conversation type are then calculated: the frequency of each conversation type, the average customer response time, the response rate, and the closing rate. The average response time, response rate, and closing rate of first-contact conversations are then normalized. A comprehensive score is calculated and ranked for each first-contact conversation. High-scoring conversation categories and related content are output for sales personnel to use as targeted reference based on channel needs, improving service efficiency and conversion rates.

[0164] As can be seen from the above, this embodiment introduces key indicators such as response rate, transaction rate and average response time by conducting structured processing and quantitative analysis on the contact data within 24 hours after the customer comes in, and establishes a comprehensive evaluation system for welcome words and first contact words through normalization and weighted scoring, which effectively avoids the problem that traditional word optimization relies on subjective experience and is difficult to quantify. By analyzing the effects of welcome words and contact words in different channels, the language structure and expression pattern with high scores are automatically identified, and users are guided to send more attractive and conversion-capable words, thereby improving the customer's response rate and service transaction rate, and realizing refined operations. This embodiment uses large language models (such as Qwen, DeepSeek, etc.) to perform language structure refinement and word classification, which can extract core expression elements and potential categories from unstructured employee language, thereby promoting the intelligent classification and continuous optimization of contact words.

[0165] This embodiment regularly analyzes and updates the scores of incoming call data, dynamically adjusts the evaluation criteria based on the performance of different channels, and realizes the continuous update and intelligent recommendation of welcome words and first contact scripts, thereby improving the overall service level. By flexibly setting the scoring weight parameters, such as focusing on the response rate, transaction rate or response experience, it supports automatic optimization of the scoring system according to the actual business scenario to meet the needs of script optimization under different operating strategies. High-scoring script expression models and contact classifications are automatically prompted in the business system to assist employees in accurately selecting or adjusting script content according to customers from different channels, thereby improving communication efficiency and customer experience. From customer access, reply, transaction to script optimization, a complete closed loop from data collection, analysis, evaluation to execution has been built, providing reliable support for intelligent services in the investment advisory industry.

[0166] In summary, the intelligent recommendation method for welcome words and first-time contact scripts described in the embodiments of the present application supports automatic optimization of the scoring system according to actual business scenarios, meeting the script optimization needs under different operating strategies. High-scoring script expression models and contact classifications are automatically prompted in the business system to assist employees in accurately selecting or adjusting script content based on customers from different channels, thereby improving communication efficiency and customer experience. From customer access, reply, transaction to script optimization, a complete closed loop from data collection, analysis, evaluation to execution has been constructed, providing reliable support for intelligent services in the investment advisory industry.

[0167] The scope of protection of the intelligent recommendation method for welcome words and first contact dialogue described in the embodiment of this application is not limited to the execution order of the steps listed in this embodiment. All solutions implemented by adding, reducing, or replacing steps in the existing technology based on the principles of this application are included in the scope of protection of this application.

[0168] According to the method provided in the embodiments of the present application, the embodiments of the present application also provide a computer program product, which includes: computer program code, which, when running on a computer, enables the computer to execute the intelligent recommendation method for welcome words and first contact scripts provided in any embodiment of the present application.

[0169] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program implements the intelligent recommendation method for welcome words and first contact scripts provided in any embodiment of the present application.

[0170] In the embodiment of the present application, any combination of one or more storage media can be used. The storage medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a RAM, a ROM, an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by an instruction execution system, device or device or used in combination with it.

[0171] An embodiment of the present application also provides an electronic device. Figure 8Shown is a structural diagram of the electronic device 100 provided in an embodiment of the present application. In some embodiments, the electronic device can be a mobile phone, a tablet computer, a wearable device, a vehicle-mounted device, an augmented reality (AR) / virtual reality (VR) device, a laptop computer, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA) and other terminal devices. In addition, the intelligent recommendation method for welcome words and first contact words provided in this application can also be applied to databases, servers and service response systems based on terminal artificial intelligence. The embodiment of this application does not impose any restrictions on the specific application scenarios of the intelligent recommendation method for welcome words and first contact words.

[0172] like Figure 8 As shown, the electronic device 100 provided in an embodiment of the present application includes a memory 101 and a processor 102 .

[0173] The memory 101 is used to store computer programs; preferably, the memory 101 includes: ROM, RAM, magnetic disk, USB flash drive, memory card or optical disk, etc., various media that can store program codes.

[0174] Specifically, the memory 101 may include a computer system readable medium in the form of a volatile memory, such as a random access memory (RAM) and / or a cache memory. The electronic device 100 may further include other removable / non-removable, volatile / non-volatile computer system storage media. The memory 101 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present application.

[0175] The processor 102 is connected to the memory 101 and is used to execute the computer program stored in the memory 101 so that the electronic device 100 executes the intelligent recommendation method for welcome words and first contact words provided in any embodiment of the present application.

[0176] Optionally, the processor 102 can be a general-purpose 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 gate or transistor logic devices, discrete hardware components.

[0177] Optionally, the electronic device 100 in this embodiment may further include a display 103. The display 103 is communicatively connected to the memory 101 and the processor 102, and is configured to display a GUI interaction interface related to the intelligent recommendation method for the welcome message and the first contact speech script.

[0178] In summary, this application achieves efficient and accurate intelligent recommendation of welcome messages and first-time contact scripts, improving overall service levels and effectively avoiding the problem of traditional script optimization relying on subjective experience and being difficult to quantify. Therefore, this application effectively overcomes the various shortcomings of the existing technology and has high industrial application value.

[0179] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical concepts disclosed in this application shall be covered by the claims of this application.

Claims

1. An intelligent recommendation method for welcome words and first contact scripts, characterized by: include: Obtain historical chat data, including historical welcome message data and historical first contact dialogue data for incoming calls from different channels; In response to accessing the chat channel, obtaining, based on the historical welcome message data, an average reply time, a reply rate, and a transaction rate for each welcome message, and obtaining a comprehensive score for each welcome message based on the average reply time, the reply rate, and the transaction rate; Obtaining high-scoring welcome messages based on the comprehensive scores of each of the welcome messages, performing language structure analysis on the high-scoring welcome messages to obtain expression patterns of the high-scoring welcome messages, and generating a welcome message corresponding to the current channel based on the expression patterns of the high-scoring welcome messages; Upon receiving chat data of a user's reply to the welcome message, obtaining the average reply time, reply rate, and transaction rate of each type of first contact script based on the historical first contact script data, and obtaining a comprehensive score for each type of first contact script based on the average reply time, reply rate, and transaction rate of each type of first contact script; According to the comprehensive score of each type of first contact script, the category and script content corresponding to the high-scoring first contact script are obtained, and the category and script content corresponding to the high-scoring first contact script are recommended to the user, so that the user can edit the first contact script corresponding to the current channel according to the category and script content corresponding to the high-scoring first contact script.

2. The intelligent recommendation method for welcome words and first contact words according to claim 1 is characterized in that: The average response time for each welcome message is obtained in the following ways: Among them, T i is the average reply time for each welcome message, T is a non-empty reply time set, min(T) is the shortest reply time, and max(T) is the longest reply time; The response rate of each welcome message is obtained in the following ways: Among them, R i is the response rate of each welcome message, C i N_C is the total count of each greeting message. i The number of greeting messages whose reply content is not empty; The methods for obtaining the conversion rate of each welcome message include: Among them, O i is the transaction rate of each welcome message, O_C i The number of service transactions for each welcome message is the number of service transaction times that are not empty.

3. The intelligent recommendation method for welcome words and first contact words according to claim 2 is characterized in that: The comprehensive score of each welcome message is obtained as follows: Normalize the average reply time, the reply rate, and the transaction rate of each welcome message to obtain the normalized average reply time, the normalized reply rate, and the normalized transaction rate; Configuring corresponding weight coefficients for the normalized average response time, the normalized response rate, and the normalized transaction rate, respectively; wherein the sum of the weight coefficients is 1; The comprehensive score of each welcome message is the sum of the normalized average reply time, the normalized reply rate, and the normalized transaction rate multiplied by their respective weight coefficients.

4. The intelligent recommendation method for welcome words and first contact words according to claim 1 is characterized in that: The performing language structure analysis on the high-scoring welcome message to obtain the expression pattern of the high-scoring welcome message includes: Accept the input welcome analysis prompt words; The AI ​​big model is called, and the AI ​​big model performs language structure analysis on the high-scoring welcome word according to the analysis requirements of the welcome word analysis prompt words to obtain the expression pattern of the high-scoring welcome word.

5. The intelligent recommendation method for welcome words and first contact words according to claim 1 is characterized in that: The average response time, response rate, and transaction rate of each type of first contact speech script obtained based on the historical first contact speech script data include: Accept input for first contact analysis prompts; Invoking an AI big model, wherein the AI ​​big model classifies the historical first contact speech data according to the analysis requirements of the first contact speech analysis prompt words to obtain the category of the first contact speech; Get the average response time, response rate, and closing rate for each type of first contact script.

6. The intelligent recommendation method for welcome words and first contact words according to claim 5 is characterized in that: The average response time for each type of first contact script is obtained in the following ways: Among them, T i is the average response time for each type of first contact script, T is a non-empty set of response times, min(T) is the shortest response time, and max(T) is the longest response time; The response rate for each type of first contact script is obtained in the following ways: Among them, R i is the response rate of each type of first contact script, C i N_C is the total number of first contact scripts for each type. i The number of first contact scripts of each type whose response content is not empty; The methods for obtaining the closing rate of each type of first contact script include: Among them, O i is the closing rate of each type of first contact script, O_C i The number of service transactions for each type of first contact script is the number of service transaction times that are not empty.

7. The intelligent recommendation method for welcome words and first contact words according to claim 6 is characterized in that: Obtaining a comprehensive score for each of the aforementioned first contact scripts includes: Normalizing the average response time, response rate, and transaction rate for each type of first contact script to obtain the normalized average response time, normalized response rate, and normalized transaction rate; Configuring corresponding weight coefficients for the normalized average response time, the normalized response rate, and the normalized transaction rate, respectively; wherein the sum of the weight coefficients is 1; The comprehensive score of each type of first contact script is the sum of the normalized average response time, the normalized response rate, and the normalized transaction rate multiplied by their respective weight coefficients.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for intelligently recommending welcome words and first-contact dialogue according to any one of claims 1 to 7 is implemented.

9. A computer program product, characterized in that The computer program product includes computer program code, which, when executed on a computer, enables the computer to implement the intelligent recommendation method for welcome words and first contact dialogue as described in any one of claims 1 to 7.

10. An electronic device, characterized in that: The electronic device comprises: processor and memory; The memory stores program instructions; The processor is configured to run the program instructions to execute the intelligent recommendation method for welcome words and first contact dialogue according to any one of claims 1 to 7.

Images