Telephone sales auxiliary system and method based on voice analysis and large model

Abstract

The invention discloses a telemarketing auxiliary system and method based on voice analysis and a large model, relates to the technical field of artificial intelligence, and aims to solve the problems of low efficiency, inconsistent communication quality and lack of professional auxiliary tools of traditional telemarketing. The method comprises the following steps: collecting sound recording files of sales and customer telephone chat which fulfill a sales goal in the field of psychological consultation, transcribing the sound recording files into text data, performing structuralization and formatting processing, and inputting the structuralized and formatted text data into a model for semantic understanding training; and generating a sales prompt document by the trained model, performing document segmentation and vectorization processing, and finally inputting a vector into the trained large model to generate a sales process suggestion. According to the invention, sales process suggestions related to psychological problems proposed by customers can be obtained in real time, and the sales efficiency and communication quality are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a telephone sales assistance system and method based on speech analysis and large models. Background Art

[0002] As stress levels in modern society continue to intensify, mental health issues are becoming increasingly prominent, leading to a growing demand for psychological counseling. However, traditional psychological counseling methods are limited by time and location, making it difficult to immediately meet the public's counseling needs. Therefore, telephone sales have become a crucial way for professional organizations to provide consulting services and establish connections with clients. However, traditional psychological counseling telephone sales processes often rely on personal experience and intuition to communicate with clients and analyze their needs. This not only leads to low efficiency but also makes it difficult to ensure consistent communication quality and sales effectiveness. Furthermore, when faced with complex or unique psychological needs, salespeople may miss opportunities to provide appropriate counseling services due to a lack of professional support tools. While some existing sales assistance systems can provide sales pitches or suggestions, these recommendations are often based on fixed templates or rules and lack dynamic adaptation and optimization to specific customer situations. In practice, salespeople still need to spend considerable time and effort analyzing customer psychology and customizing strategies, resulting in limited improvements in sales efficiency. Summary of the Invention

[0003] The object of the present invention is to provide a telephone sales assistance system and method based on speech analysis and large models to solve the problems raised in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a telephone sales assistance method based on speech analysis and a large model, the telephone sales assistance method comprising:

[0005] Step S1: Collect all audio recordings of salespeople who have achieved sales targets and had phone conversations with customers;

[0006] Step S2: transcribing the audio recording file into text data, and performing data cleaning on the text data to convert it into structured text data;

[0007] Step S3: formatting the structured text data and inputting it into the model for semantic understanding training;

[0008] Step S4: The trained model automatically generates a sales prompt document, and performs document segmentation and vectorization on the sales prompt document;

[0009] Step S5: Input the vector into the trained model to generate sales process recommendations.

[0010] Furthermore, the process of transcribing the recording file into text data in step S2 includes:

[0011] Step S2-1-1: Segment each recording file into segments according to the conversation turns between the salesperson and the customer;

[0012] Step S2-1-2: using automatic speech recognition technology to transcribe each of the segments into text data;

[0013] Step S2-1-3: performing quality inspection on the text data to correct transcription errors;

[0014] The above steps are the first step in converting audio recordings into structured text data. By segmenting long recordings according to the conversation turns between sales and customers, transcription efficiency is improved and the clarity of the context is ensured. Automatic speech recognition technology is then used to transcribe each segment to generate text data. Finally, the transcription results are quality checked to correct obvious transcription errors and ensure that the content is complete and consistent with the actual call content.

[0015] Furthermore, the process of cleaning the text data in step S2 includes:

[0016] Step S2-2-1: removing pause words in the text data through a cleaning script;

[0017] The above steps are designed to remove irrelevant pause words and repetitive expressions, such as "hmm," "oh," and "ah," which are common catchphrases and interjections. These words frequently appear in client counseling conversations but are unhelpful for semantic understanding. Regular expressions or cleaning scripts can be used to remove them in batches to ensure the conciseness and clarity of the text.

[0018] Step S2-2-2: adding or modifying punctuation marks in the text data according to the dialogue logic and grammatical conventions;

[0019] The above steps are designed to make the text data more suitable for reading and analysis. This step helps improve the semantic understanding of sentences by the large model and prevents the impact of missing or misused punctuation on model training results.

[0020] Step S2-2-3: using a spell checker to detect and correct spelling errors in the text data;

[0021] The above steps are designed to detect and correct spelling errors in text data. Speech recognition may make mistakes, especially when affected by accents or background noise. This step ensures data consistency and improves the accuracy of the large model in recognizing input text.

[0022] Step S2-2-4: removing non-dialog annotations from the text data using automatic script recognition technology;

[0023] The above steps are intended to filter out information irrelevant to the core content of the conversation, such as background noise, non-conversational instructions, or comments. These contents can be identified through automatic scripting, and unnecessary parts can be removed to ensure that the model focuses on the semantics and sentiment of the conversation itself.

[0024] Furthermore, the process of inputting the formatted text data into the model for training in step S3 includes:

[0025] Step S3-1: loading the formatted text data into a training framework;

[0026] The above steps are to load the dataset and load the formatted and structured text data into the training framework. For training frameworks such as TensorFlow and PyTorch, the text data is preprocessed to meet the input and output requirements of the model so that it can be effectively understood and processed by the model.

[0027] Step S3-2: Select a pre-trained model and load the model;

[0028] The above steps involve selecting a suitable large model architecture, such as GPT and BERT, and loading pre-trained basic models. Leveraging the language understanding capabilities of these pre-trained models, the models are then adapted through transfer learning to assist telephone sales in the field of psychological counseling.

[0029] Step S3-3: loading pre-trained weights during the initialization phase of the model;

[0030] The above steps load pre-trained weights during the model initialization phase, giving the model general language expression capabilities. Through this step, the model quickly acquires basic semantic understanding and is fine-tuned based on the specific data of sales conversations, gradually adapting to the semantic requirements of the psychological counseling field. This process allows the model to transition from general understanding to a specific understanding of telephone sales assistance.

[0031] Step S3-4: repeatedly inputting the formatted text data into the model for training using a supervised learning algorithm, and measuring the difference between the model prediction and the true label using a loss function until the difference reaches a threshold;

[0032] During the fine-tuning phase, the above steps use supervised learning algorithms and multiple models for targeted training. By repeatedly inputting the above text data, the model gradually improves its semantic understanding of specific situations and its ability to generate accurate responses. Loss functions such as cross-entropy are used to measure the gap between the model's predictions and the true labels, thereby guiding the optimization of model parameters.

[0033] Step S3-5: pre-processing the newly generated recording file of the telephone conversation between the salesperson who achieved the sales target and the customer through a recording device and converting it into formatted text data, and adding the formatted text data to the model for retraining;

[0034] After fine-tuning the model, the above steps are optimized through continuous feedback, and the model performance is monitored in real time in actual conversations. Newly generated telephone chat recordings that have achieved sales targets are added to the training dataset after processing. The model is retrained or fine-tuned regularly to enhance the model's adaptability and generation quality, so that it maintains efficient understanding and responsiveness in the ever-changing psychological counseling conversation scenarios.

[0035] Furthermore, the step of segmenting and vectorizing the sales prompt document in step S4 includes:

[0036] Step S4-1: Identify the subject and content of each paragraph in the sales tip document using a semantic analysis tool;

[0037] The sales tips documents mentioned in the previous step include sales process documents and sales method elements documents. These documents are analyzed to identify the theme and content of each paragraph. To ensure the semantic integrity of the segmented segments, reasonable segmentation rules are established, such as segmentation by logical segments or semantic coherence. Natural language processing techniques such as sentence segmentation models or dependency parsing can be used to find natural demarcation points, so that each segment has the characteristics of an independent semantic unit.

[0038] Step S4-2: dividing the sales tip document into text segments according to the subject and content;

[0039] Step S4-3: using an embedding model to capture the semantic features of the text segment, representing the text segment as a vector, and storing the vector in a vector database;

[0040] The above steps use embedding models such as Word2Vec and BERT to convert the segmented text fragments into high-dimensional vectors. The embedding model captures the semantic features of the text and represents each fragment as a vector for semantic similarity calculation. After vectorization, the text is no longer a simple character or phrase, but is converted into a point in the vector space. In this space, the distance between vectors reflects the semantic similarity. The vectors of all text fragments are stored in a vector database, allowing the system to quickly retrieve relevant content based on semantic similarity, thereby improving retrieval accuracy and ensuring that the provided content has a higher degree of semantic match with the customer's question.

[0041] Furthermore, the segmentation method in step S4-2 is to determine a cutoff point based on the subject and content of each paragraph in the sales tip document using a sentence model, and to divide the sales tip document into semantically coherent text segments based on the cutoff point, wherein each of the text segments includes a subject tag and a keyword, and the length of the text segment is limited according to a set threshold;

[0042] The above steps divide the document into text segments according to the set semantic logic. Each segment is independent and complete in meaning, so that relevant responses can be provided quickly when customers ask questions. To optimize retrieval performance, each segment should contain core topic tags and keywords, and maintain an appropriate segment length based on the set threshold to avoid being too short or too long to affect the understanding effect, thereby ensuring that the segment has the ability to express independently and be self-sufficient in information.

[0043] In order to better implement the above method, the present invention also provides a telephone sales assistance system based on voice analysis and large models, the telephone sales assistance system includes: a Bluetooth module, a transmission module, an application terminal, and a cloud server;

[0044] Bluetooth module, used to receive recordings of telephone conversations between sales and customers;

[0045] A transmission module, used for transmitting the recording file and sales process suggestions;

[0046] The application side is used to receive the recording file and upload the recording file to the cloud server, and receive the sales process suggestions returned by the cloud server;

[0047] The cloud server is used to receive the recording file and process the recording file to generate sales process suggestions.

[0048] Furthermore, the cloud server includes a speech recognition module, a structuring and formatting module, a model training module, and a document processing module;

[0049] A speech recognition module, which divides the recording file into segments and transcribes each segment into text data;

[0050] A structuring and formatting module, configured to clean the text data to convert the text data into structured text data, and format the structured text data into formatted text data;

[0051] A model training module, inputting the formatted text data into a model to train the model so that the model can automatically generate a sales reminder document;

[0052] The document processing module is used to segment and vectorize the sales prompt document and store the vector.

[0053] Furthermore, the document processing module includes a document segmentation unit, a vectorization unit, a vector storage unit, and a sales process suggestion generation unit;

[0054] a document segmentation unit for identifying the subject and content of each paragraph in the sales tip document and segmenting the sales tip document into text segments based on the subject and content;

[0055] A vectorization unit, which captures the semantic features of the text segment through an embedding model and represents the text segment as a vector;

[0056] A vector storage unit, used for storing vectors generated by the vectorization unit into a vector database;

[0057] The sales process suggestion generating unit searches the vector database according to a similarity threshold to obtain a vector with the highest similarity, and inputs the vector with the highest similarity into the trained model to generate a sales process suggestion.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. Improve sales efficiency and communication quality

[0060] Automatic speech recognition technology converts recorded phone conversations between sales representatives and customers into text data. This data is then cleaned and structured, enabling sales representatives to accurately capture the psychological issues customers are seeking advice on. Furthermore, a large model is used to perform semantic understanding training on formatted text data, automatically generating sales tips documents that provide sales staff with timely and accurate sales support information. This process not only significantly improves sales staff efficiency but also ensures consistency in communication quality and sales effectiveness.

[0061] 2. Dynamically adjust and optimize sales strategies

[0062] This system dynamically adjusts and optimizes sales strategies based on specific customer counseling needs. By segmenting and vectorizing sales document segments, the system can quickly retrieve historical counseling cases from a vector database that most closely resemble the customer's problem, providing sales staff with targeted sales process recommendations. This dynamic adjustment capability based on specific customer counseling needs enables sales staff to more accurately grasp customer needs and improve sales success rates.

[0063] 3. Enhance the professional capabilities of sales staff

[0064] The system also features continuous learning and optimization capabilities. By continuously collecting new sales recordings and incorporating them into the model for retraining, the system continuously improves its semantic understanding and sales support capabilities. This enables salespeople to maintain efficient understanding and responsiveness in the ever-changing landscape of psychological counseling conversations, thereby continuously enhancing their professional capabilities.

[0065] 4. Reduce human errors and costs

[0066] In traditional telephone sales, salespeople rely heavily on personal experience and intuition to communicate with customers and analyze their psychological counseling needs. This often leads to low efficiency and difficulty ensuring consistency. This invention, however, uses automation and intelligent methods to accurately capture and analyze sales conversations, significantly reducing human errors and costs. Furthermore, the system can provide salespeople with standardized sales process recommendations, further reducing sales discrepancies caused by human factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 Schematic diagram of the method of the present invention;

[0068] Figure 2 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION

[0069] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0070] Example: Figure 1-Figure 2 As shown, the present invention provides a technical solution, a telephone sales assistance method based on speech analysis and a large model, the telephone sales assistance method comprising:

[0071] Step S1: Collect all audio recordings of sales calls with customers that have achieved sales targets;

[0072] Step S2: transcribing the audio recording file into text data, and performing data cleaning on the text data to convert it into structured text data;

[0073] The process of transcribing the recording file into text data in step S2 includes:

[0074] Step S2-1-1: Segment each recording file into segments according to the conversation turns between the salesperson and the customer;

[0075] Step S2-1-2: using automatic speech recognition technology to transcribe each of the segments into text data;

[0076] Step S2-1-3: Perform quality inspection on the text data to correct transcription errors.

[0077] The process of cleaning the text data in step S2 includes:

[0078] Step S2-2-1: removing pause words in the text data through a cleaning script;

[0079] Step S2-2-2: adding or modifying punctuation marks in the text data according to the dialogue logic and grammatical conventions;

[0080] Step S2-2-3: using a spell checker to detect and correct spelling errors in the text data;

[0081] Step S2-2-4: Remove the non-dialog annotation part in the text data through automatic script recognition technology.

[0082] Step S3: formatting the structured text data and inputting it into the model for semantic understanding training;

[0083] The process of inputting the formatted text data into the model for training in step S3 includes:

[0084] Step S3-1: loading the formatted text data into a training framework;

[0085] Step S3-2: Select a pre-trained model and load the model;

[0086] Step S3-3: loading pre-trained weights during the initialization phase of the model;

[0087] Step S3-4: repeatedly inputting the formatted text data into the model for training using a supervised learning algorithm, and measuring the difference between the model prediction and the true label using a loss function until the difference reaches a threshold;

[0088] Step S3-5: pre-process the newly generated recording file of the telephone conversation between the salesperson who has achieved the sales target and the customer through a recording device and convert it into formatted text data, and add the formatted text data to the model for retraining.

[0089] Step S4: The trained model automatically generates a sales prompt document, and performs document segmentation and vectorization on the sales prompt document;

[0090] The step of segmenting and vectorizing the sales prompt document in step S4 includes:

[0091] Step S4-1: Identify the subject and content of each paragraph in the sales tip document using a semantic analysis tool;

[0092] Step S4-2: dividing the sales tip document into text segments according to the subject and content;

[0093] Step S4-3: Using an embedding model to capture the semantic features of the text segment, representing the text segment as a vector, and storing the vector in a vector database.

[0094] Among them, the segmentation method of the fragment in step S4-2 is that the sentence model determines the dividing point according to the theme and content of each paragraph in the sales tip document, and divides the sales tip document into semantically coherent text fragments according to the dividing point, wherein each of the text fragments includes a subject tag and keywords, and the length of the text fragment is limited according to the set threshold.

[0095] Step S5: Input the vector into the trained model to generate sales process recommendations.

[0096] In order to better implement the above method, the present invention also provides a telephone sales assistance system based on voice analysis and large models, the telephone sales assistance system includes: a Bluetooth module, a transmission module, an application terminal, and a cloud server;

[0097] Bluetooth module, used to receive recordings of telephone conversations between sales and customers;

[0098] A transmission module, used for transmitting the recording file and sales process suggestions;

[0099] The application side is used to receive the recording file and upload the recording file to the cloud server, and receive the sales process suggestions returned by the cloud server;

[0100] The cloud server is used to receive the recording file and process the recording file to generate sales process suggestions.

[0101] In an embodiment of the present invention, the application connects to a Bluetooth module and wireless headset. When a salesperson speaks with a customer through the headset, the audio from both sides of the call is transmitted to the application in real time. The application continuously transmits the audio stream to a cloud server via the computer or mobile phone network. Upon receiving the audio stream, the cloud server performs real-time speech recognition and conversion, converting the speech into text, and then sends the text to the cloud server's large model for processing.

[0102] Among them, the cloud server includes a speech recognition module, a structuring and formatting module, a model training module, and a document processing module;

[0103] A speech recognition module, which divides the recording file into segments and transcribes each segment into text data;

[0104] A structuring and formatting module, configured to clean the text data to convert the text data into structured text data, and format the structured text data into formatted text data;

[0105] A model training module, inputting the formatted text data into a model to train the model so that the model can automatically generate a sales reminder document;

[0106] The document processing module is used to segment and vectorize the sales prompt document and store the vector.

[0107] The document processing module includes a document segmentation unit, a vectorization unit, a vector storage unit, and a sales process suggestion generation unit;

[0108] a document segmentation unit for identifying the subject and content of each paragraph in the sales tip document and segmenting the sales tip document into text segments based on the subject and content;

[0109] A vectorization unit, which captures the semantic features of the text segment through an embedding model and represents the text segment as a vector;

[0110] A vector storage unit, used for storing vectors generated by the vectorization unit into a vector database;

[0111] The sales process suggestion generating unit searches the vector database according to a similarity threshold to obtain a vector with the highest similarity, and inputs the vector with the highest similarity into the trained model to generate a sales process suggestion.

[0112] In an embodiment of the present invention, the big model of the cloud server performs semantic understanding of text information, converts the text information into vectors, searches the vector database for the top five vectorized fragments with the highest similarity, and submits these fragments to the big model for integration and summary through the Langchain method; combines the sales process library and method library content in the knowledge base with the semantic context of the actual conversation to generate appropriate sales process suggestions; after the application end receives the sales process suggestions returned by the cloud, it is displayed on the application interface, and the process suggestions will be updated every 15 seconds on the application interface for reference by sales personnel.

[0113] In an embodiment of the present invention, the sales process displayed on the application interface is suggested as follows:

[0114] (1) Please continue to ask the client's basic information, including his emotional experience.

[0115] (2) Ask the customer what is the cause of their current troubles? What is the trigger?

[0116] (3) Based on the client’s narration, we can preliminarily judge that his / her original family is a doting family. The characteristics and behavior patterns of the doting family can be explained to the client.

[0117] (4) According to the client’s description, he has a typical borderline personality disorder. Please analyze this personality disorder and give him some suggestions.

[0118] (5) If the customer shows distrust in you, please communicate more with the customer, show more empathy, and demonstrate your professional level to gain the customer's trust. Since the customer's trust level is low, they cannot fully accept your analysis and guidance, which is not conducive to solving the problem. You can gradually improve the customer's trust through further communication and empathy.

[0119] (6) All sales processes have been carried out smoothly and the conversation has covered all the key elements of the sales method; now is a good time to recommend psychological counseling to him. You can end the current topic in time, recommend psychological counseling services to him and provide a quotation.

[0120] (7) If the customer has doubts about the service, mainly about the price, please help the customer resolve the price doubts by demonstrating the unique value and irreplaceability of the service. You can explain the rationality behind the price in combination with the long-term benefits brought by the service.

[0121] (8) If the customer accepts the service price you recommended, please proceed to the signing step immediately.

[0122] In an embodiment of the present invention, the above-mentioned sales process suggestions are updated every 15 seconds during the entire call process, and the updated sales process suggestions are displayed on the salesperson's computer or mobile phone screen, thereby improving the salesperson's efficiency in providing psychological counseling and sales to customers and promoting the achievement of sales.

[0123] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A telephone sales assistance method based on speech analysis and large models, characterized in that: The telephone sales assistance method comprises: Step S1: Collect all the recordings of telephone conversations between salespeople who have achieved sales targets and their customers; Step S2: transcribing the recording file into text data, and performing data cleaning on the text data and converting it into structured text data; Step S3: formatting the structured text data and inputting it into the model for semantic understanding training; Step S4: the trained model automatically generates a sales reminder document, and performs document segmentation and vectorization on the sales reminder document; Step S5: Input the vector into the trained model to generate sales process recommendations.

2. The telephone sales assistance method based on speech analysis and large models according to claim 1 is characterized by: The process of transcribing the recording file into text data in step S2 includes: Step S2-1-1: Each recording file is segmented according to the conversation turns between the salesperson and the customer; Step S2-1-2: using automatic speech recognition technology to transcribe each of the segments into text data; Step S2-1-3: Perform quality inspection on the text data to correct transcription errors.

3. The telephone sales assistance method based on speech analysis and large models according to claim 1 is characterized in that: The process of cleaning the text data in step S2 includes: Step S2-2-1: removing pause words in the text data by using a cleaning script; Step S2-2-2: adding or modifying punctuation marks in the text data according to the dialogue logic and grammatical habits; Step S2-2-3: using a spelling check tool to detect and correct spelling errors in the text data; Step S2-2-4: Remove the non-dialog annotation part in the text data through automatic script recognition technology.

4. The telephone sales assistance method based on speech analysis and large models according to claim 1 is characterized by: The process of inputting the formatted text data into the model for training in step S3 includes: Step S3-1: loading the formatted text data into a training framework; Step S3-2: Select a pre-trained model and load the model; Step S3-3: loading pre-trained weights during the initialization phase of the model; Step S3-4: repeatedly inputting the formatted text data into the model for training using a supervised learning algorithm, and measuring the difference between the model prediction and the true label using a loss function until the difference reaches a threshold; Step S3-5: pre-process the newly generated recording file of the telephone chat between the salesperson who has achieved the sales target and the customer and convert it into formatted text data through a recording device, and add the formatted text data to the model for retraining.

5. The telephone sales assistance method based on speech analysis and large models according to claim 1 is characterized by: The step of segmenting and vectorizing the sales prompt document in step S4 includes: Step S4-1: Identify the subject and content of each paragraph in the sales tip document by using a semantic analysis tool; Step S4-2: dividing the sales reminder document into text segments according to the subject and content; Step S4-3: using an embedding model to capture the semantic features of the text segment, representing the text segment as a vector, and storing the vector in a vector database.

6. The telephone sales assistance method based on speech analysis and large models according to claim 5 is characterized by: The method for segmenting the fragments in step S4-2 is that a sentence model determines a dividing point according to the theme and content of each paragraph in the sales tip document, and divides the sales tip document into semantically coherent text fragments according to the dividing point, wherein each of the text fragments includes a topic tag and keywords, and the length of the text fragment is limited according to a set threshold.

7. A telephone sales assistance system based on speech analysis and large models, used to execute a telephone sales assistance method based on speech analysis and large models as claimed in any one of claims 1 to 6, characterized in that: The telephone sales assistance system includes: a Bluetooth module, a transmission module, an application terminal, and a cloud server; The Bluetooth module is used to collect audio files of telephone conversations between sales and customers; The transmission module is used to transmit the recording file and the sales process suggestion; The application end is used to receive the recording file and upload the recording file to the cloud server, and receive the sales process suggestion returned by the cloud server; The cloud server is used to receive the recording file and process the recording file to generate sales process suggestions.

8. The telephone sales assistance system based on speech analysis and large models according to claim 7 is characterized by: The cloud server includes a speech recognition module, a structuring and formatting module, a model training module, and a document processing module; The speech recognition module divides the recording file into segments and transcribes each segment into text data; The structuring and formatting module is used to clean the text data to convert the text data into structured text data, and format the structured text data into formatted text data; The model training module inputs the formatted text data into a model to train the model so that the model can automatically generate a sales reminder document; The document processing module is used to segment and vectorize the sales prompt document and store the vector.

9. The telephone sales assistance system based on speech analysis and large models according to claim 7, characterized in that: The document processing module includes a document segmentation unit, a vectorization unit, a vector storage unit, and a sales process suggestion generation unit; The document segmentation unit identifies the subject and content of each paragraph in the sales tip document, and segments the sales tip document into text segments according to the subject and content; The vectorization unit captures the semantic features of the text segment through an embedding model and represents the text segment as a vector; The vector storage unit is used to store the vector generated by the vectorization unit in a vector database; The sales process suggestion generating unit queries the vector database according to the similarity threshold to obtain the vector with the highest similarity, and inputs the vector with the highest similarity into the trained model to generate the sales process suggestion.

Images