User intention recognition method and device based on artificial intelligence, equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a user intention recognition method and device based on artificial intelligence, equipment and a medium, and the method comprises the steps: obtaining user search information, and generating a document topic model according to the user search information; performing text preprocessing on the user search information through the document topic model to obtain optimized search data; inputting the optimized search data into an intention recognition model to obtain intention classification; and generating and displaying message prompts according to the intention classification. According to the method and the system, the real demands and intentions of the user can be accurately captured by analyzing the short text search behavior of the user, and in the seat operation process, the system can provide personalized prompt information according to the data to help the seat to quickly understand the customer demands, so that the communication efficiency and the operation efficiency are remarkably improved, and the user experience is improved. The intelligent assistance not only reduces the workload of the seat, but also improves the customer satisfaction and promotes the improvement of the sales conversion rate.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence technology and natural language processing technology, and in particular to an artificial intelligence-based user intention recognition method, device, equipment and medium. Background Art

[0002] In the current company's auto insurance business, efficient processing of the breakpoint list has become the key to improving customer satisfaction and conversion rate. The breakpoint list refers to the list of potential customers who actively search online and show high purchase intention. However, in the actual operation process, the seats are faced with a series of challenges, which affect the efficiency and effectiveness of the operation.

[0003] First, a large number of breakpoint lists are sent to the agents every day, making their workload extremely heavy. This not only increases work pressure, but may also lead to a decline in service quality, because the agents may not be able to pay enough attention to and follow up on each potential customer.

[0004] Secondly, after obtaining the breakpoint list, agents often lack personalized intent prompts about these lists. This means that they need to spend extra time and energy to explore and understand customers' needs and preferences during the communication process. This uncertainty not only reduces communication efficiency, but may also affect customer experience and satisfaction. Summary of the invention

[0005] The present invention provides an artificial intelligence user intention recognition method, device, computer equipment and medium, which can obtain the user's real intention through short text search of the user, and then give prompts during the agent's operation to improve the agent's operation efficiency.

[0006] In a first aspect, a method for user intention recognition and processing based on artificial intelligence is provided, comprising:

[0007] Obtaining user search information, and generating a document topic model according to the user search information;

[0008] Performing text preprocessing on the user search information through the document topic model to obtain optimized search data;

[0009] Inputting the optimized search data into an intent recognition model to obtain intent classification;

[0010] Generate and display a message prompt based on the intent classification.

[0011] In a second aspect, an intelligent question-answering processing device based on artificial intelligence is provided, comprising:

[0012] An acquisition module, used to acquire user search information and generate a document topic model according to the user search information;

[0013] A processing module, used for performing text preprocessing on the user search information through the document topic model to obtain optimized search data;

[0014] An input module, used for inputting the optimized search data into an intent recognition model to obtain an intent classification;

[0015] A generation module is used to generate and display message prompts according to the intent classification.

[0016] In a third aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned artificial intelligence-based user intent recognition method when executing the computer program.

[0017] In a fourth aspect, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned artificial intelligence-based user intention recognition method are implemented.

[0018] In the scheme implemented by the above-mentioned artificial intelligence-based user intention recognition method, device, computer equipment and storage medium, user search information can be obtained through the client, and a document topic model can be generated based on the user search information; the user search information is pre-processed through the document topic model to obtain optimized search data; the optimized search data is input into the intention recognition model to obtain intent classification; and message prompts are generated and displayed based on the intent classification. In the present invention, for complex insurance entities such as the company's auto insurance business, the user's real needs and intentions can be accurately captured by analyzing the user's short text search behavior. During the agent operation process, the system can provide personalized prompt information based on these data to help the agent quickly understand customer needs, thereby significantly improving communication efficiency and operation efficiency. This intelligent assistance not only reduces the workload of the agent, but also improves customer satisfaction and promotes the improvement of sales conversion rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0020] Figure 1 is a schematic diagram of an application environment of a method for identifying user intention based on artificial intelligence in one embodiment of the present invention;

[0021] Figure 2is a flow chart of a method for identifying user intention based on artificial intelligence in one embodiment of the present invention;

[0022] Figure 3 yes Figure 1 A schematic flow chart of a specific implementation of step S100;

[0023] Figure 4 yes Figure 1 A schematic flow chart of a specific implementation of step S200;

[0024] Figure 5 yes Figure 4 A schematic flow chart of a specific implementation of step S210;

[0025] Figure 6 yes Figure 4 A schematic flow chart of a specific implementation of step S220;

[0026] Figure 7 yes Figure 4 A schematic flow chart of a specific implementation of step S230;

[0027] Figure 8 yes Figure 1 A schematic flow chart of a specific implementation of step S300;

[0028] Fig. 9 is a structural schematic diagram of a user intention recognition device based on artificial intelligence in one embodiment of the present invention;

[0029] Fig.10 is a schematic diagram of a structure of a computer device in one embodiment of the present invention;

[0030] Fig.11 It is another structural schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] The user intention recognition method based on artificial intelligence provided by the embodiment of the present invention can be applied in the following aspects: Figure 1In the application environment, the client communicates with the server through the network. The server can obtain user search information through the client, and generate a document theme model according to the user search information; perform text preprocessing on the user search information through the document theme model to obtain optimized search data; input the optimized search data into the intention recognition model to obtain the intention classification; generate and display message prompts according to the intention classification. In the present invention, for complex insurance entities such as the company's auto insurance business, the real needs and intentions of users can be accurately captured by analyzing the short text search behavior of users. During the agent operation process, the system can provide personalized prompt information based on these data to help agents quickly understand customer needs, thereby significantly improving communication efficiency and operation efficiency. This intelligent assistance not only reduces the workload of agents, but also improves customer satisfaction and promotes the improvement of sales conversion rate. Among them, the client can be but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices. The server can be implemented with an independent server or a server cluster composed of multiple servers. The present invention is described in detail below through specific embodiments.

[0033] See also Figure 2 As shown, Figure 2 A flowchart of a method for identifying user intention based on artificial intelligence provided by an embodiment of the present invention includes the following steps:

[0034] S100: Obtain user search information, and generate a document topic model according to the user search information.

[0035] The user intention recognition method based on artificial intelligence provided by the present invention can be applied to intelligent processing platforms such as intelligent customer service or intelligent assistants in various application scenarios. The intelligent processing platform is usually implemented through a server, which can obtain user search information in real time. For example, in the field of insurance applications, users often ask questions by actively searching online. The intelligent processing platform obtains the user's search information and then generates the topic type of the user's touch point search online.

[0036] For example, in the insurance industry, especially in the field of auto insurance, using user search information to optimize service processes and improve customer experience is an important technological innovation. By obtaining users' search information on the platform, we can gain a deep understanding of their needs and preferences. After analysis, this data can be used to build a document topic model that can identify key topics in user search content.

[0037] The system first captures the user's search query, which may include keywords, phrases or natural language queries, and uses the user search information to generate a document topic model. This usually involves using algorithms such as LDA (latent Dirichlet allocation) to identify the main topics in the search query.

[0038] In modern auto insurance business, the system first captures the user's search queries, which may include keywords, phrases or natural language expressions. Using this rich user search information, the system generates a document topic model through advanced algorithms such as Latent Dirichlet Allocation (LDA). This process involves identifying the main topics in the search query, thereby revealing the user's real needs and interests. Through this technical means, the system can deeply understand the user's intentions and provide targeted prompts in the subsequent agent operation process. This not only improves the operating efficiency of the agents, enabling them to respond to customer needs more quickly, but also significantly improves customer satisfaction and service quality.

[0039] In addition, this method can also help companies better analyze market trends and customer behavior, and provide data support for product development and marketing strategies. Therefore, using user search information to generate document topic models and applying them to auto insurance business not only optimizes customer service processes, but also brings deeper business insights and competitive advantages to companies.

[0040] like Figure 3 As shown, in step S100, generating a document topic model according to the user search information includes the following steps:

[0041] S110, acquiring a plurality of search topics according to the user search information;

[0042] S120: Use the search topic that appears most frequently in the user search information as the document topic model.

[0043] For steps S110-S120, in the auto insurance business, the system obtains multiple search topics by analyzing the user's search information. It should be understood that in the user's active online search log, there will be multiple different topics involved, such as topics about medical insurance, topics about natural disasters, and issues about insurance premiums, etc. These topics reflect the user's interests and needs, and provide important references for subsequent services. In order to build an effective document topic model, the system will identify the search topics that appear most frequently in all user search information. This frequently occurring search topic is selected as the core of the document topic model because it represents the main focus of most users.

[0044] In this way, the document topic model can more accurately capture the common needs of the user group, thereby providing more accurate prompts and support during the agent operation process. This data-driven approach not only improves the work efficiency of the agents, enabling them to respond to customer needs more quickly, but also significantly enhances customer satisfaction and service quality.

[0045] In addition, this method can also help enterprises better understand market dynamics and customer behavior, providing strong data support for product development and marketing strategies. Therefore, using the user search information to obtain search topics and taking the most frequently occurring search topic as the document topic model not only optimizes the customer service process but also brings deeper business insights and competitive advantages to the enterprise.

[0046] S200. Perform text preprocessing on the user search information through the document topic model to obtain optimized search data.

[0047] In the auto insurance business, preprocessing the user search information is a key step in constructing an effective document topic model. The preprocessing includes multiple stages. First is the cleaning process, which aims to remove noise and irrelevant information in the data, such as special characters, numbers, or meaningless filler words. This step ensures the accuracy and efficiency of subsequent analysis.

[0048] Next is the word segmentation step, which is the process of splitting continuous text into meaningful words or phrases. In the Chinese context, word segmentation is particularly important because Chinese writing usually does not use spaces to separate words. Accurate word segmentation is crucial for understanding the true intentions of users. Removing stop words is another key link. Stop words refer to words that frequently appear in the text but have little analytical value, such as "de" (的), "shi" (是), etc. Although these words are common, they do not carry useful information, so they need to be removed before constructing the document topic model.

[0049] In addition to the above basic steps, it may also include more advanced processing techniques such as stemming and lemmatization to further standardize the text data. Stemming is the process of reducing a word to its basic form (root), while lemmatization is the process of converting different forms of a word to its most basic form.

[0050] Through this series of preprocessing steps, the quality and consistency of the user search information can be ensured, laying a solid foundation for subsequent topic modeling and intent recognition. This refined data processing not only improves the accuracy of the document topic model but also enables agents to provide more personalized services based on clearer user intentions, thereby significantly enhancing customer satisfaction and business efficiency.

[0051] Such as Figure 4As shown, in step S200, the text preprocessing of the user search information by the document topic model to obtain optimized search data includes:

[0052] S210, performing text segmentation and text denoising processing on the user search information to obtain an optimized short text;

[0053] S220, expanding the optimized short text to obtain preliminary optimized search data;

[0054] S230, generating a personalized representation of user search through the topic vector of the document topic model and the topic distribution of the user search information;

[0055] S240: Filter and screen the preliminary optimized search data according to the user search personalized representation to obtain the optimized search data.

[0056] In the auto insurance business, first of all, the continuous text is segmented into meaningful words or phrases through text segmentation technology. This is because the accuracy of the segmentation directly affects the effect of subsequent topic analysis and intent recognition. For example, for the input "I want to know the auto insurance claim process", the segmentation technology will segment it into keywords such as "I", "want", "know", "auto insurance", "claim" and "process". These keywords can help better understand the user's needs and intentions.

[0057] Next, we perform text denoising. In this process, we need to remove noise and irrelevant information from the data, such as special characters, numbers, or meaningless filler words. The purpose of this step is to ensure the quality and consistency of the data, thereby improving the accuracy of subsequent processing. For example, if a user enters a text containing a large number of special characters or numbers, this noise information needs to be removed in order to more accurately understand the user's intention.

[0058] After these preprocessing steps, the resulting optimized short text can be further expanded to form preliminary optimized search data. This step involves matching the segmented text with a predefined thesaurus or related concept library. In this way, the text content can be enriched and its expressiveness can be improved. For example, if the user enters "car insurance", it can be matched with "car insurance" in the thesaurus to obtain a more comprehensive search result.

[0059] Subsequently, the topic vector of the document topic model and the topic distribution of the user search information are used to generate the user search personalized representation. This process involves complex algorithm calculations, aiming to capture the deep intentions and preferences behind each user's search behavior. In this way, the user's needs can be better understood and more personalized services can be provided.

[0060] Finally, according to the personalized representation of the user search, the preliminary optimized search data is filtered and screened to obtain the final optimized search data. For example, if there is data in the preliminary optimized search data that does not match the personalized representation of the user search, the data will be filtered and deleted, and the data that matches the personalized table of the user search will be saved. This filtering process is based on the personalized needs of the user to ensure that the search data provided to the seat staff is both relevant and targeted. This can not only improve work efficiency, but also improve customer satisfaction.

[0061] The entire process not only improves the accuracy of the document topic model, but also enables agents to provide more personalized services based on clearer user intent. This personalized service provision significantly improves customer satisfaction and business efficiency. The application of this method enables insurance companies to stand out in the fiercely competitive market and provide customers with higher-value products and services.

[0062] like Figure 5 As shown, in step S210, the user search information is subjected to text segmentation and text denoising to obtain an optimized short text, including:

[0063] S211, using the Viterbi algorithm to perform text segmentation on the user search information to obtain a preliminary short text;

[0064] S212, eliminating stop words in the preliminary short text, and using a word vector model to perform word vector training on the preliminary short text from which the stop words are eliminated, to obtain the optimized short text.

[0065] In the auto insurance business, in order to improve service quality and efficiency, it is first necessary to accurately process the text of user search information. Using the Viterbi algorithm to segment user search information is an effective method. The Viterbi algorithm can identify the most likely word sequence based on a statistical model, thereby segmenting continuous text into meaningful words or phrases to obtain preliminary short text.

[0066] The word vector model is a TextRNN+Attention model, which is a deep learning model that combines a recurrent neural network (RNN) and an attention mechanism for text classification tasks. This model can better capture key information in the text and improve the model's effectiveness and interpretability by adding an attention mechanism to the traditional TextRNN model.

[0067] The core of the TextRNN+Attention model lies in its combination of the advantages of the RNN and Attention mechanisms. The RNN can process sequential data and capture context information in the sequence through hidden states. The Attention mechanism, on the other hand, can assign a weight to each word to measure its contribution to the current task. In this way, the model can not only consider the overall structure of the text but also highlight the role of important words, thereby improving the accuracy of classification.

[0068] Next, removing stop words from these preliminary short texts is a necessary step. Stop words refer to words that frequently appear in the text but have little analytical value, such as "de" (of), "shi" (is), etc. Removing these stop words can reduce the noise in the data and improve the accuracy of subsequent processing. Then, use a word vector model to perform word vector training on the preliminary short texts after removing stop words. The word vector model can convert the words in the text into vector representations of a fixed length, and these vectors contain the semantic information of the words. In this way, the text data can be transformed into a numerical form that machine learning algorithms can process. The resulting optimized short texts not only remove irrelevant information but also retain rich semantic content, providing high-quality input data for constructing a document topic model.

[0069] In the field of natural language processing (NLP), text preprocessing is a crucial step. It is not just simple operations such as word segmentation and stop word removal but also includes a series of complex processes such as part-of-speech tagging and entity recognition. Part-of-speech tagging means assigning a part of speech (such as noun, verb, etc.) to each word in the text, which helps to further understand the structure and meaning of the sentence. Entity recognition is to identify proper nouns in the text, such as person names, place names, organization names, etc., which is very important for information extraction and knowledge graph construction.

[0070] In addition, text preprocessing also includes tasks such as sentiment analysis and keyword extraction. Sentiment analysis aims to determine the sentiment tendency expressed in the text, such as positive, negative, or neutral. This is particularly important in areas such as brand monitoring, public opinion analysis, and customer satisfaction surveys. Keyword extraction is to identify the most important words or phrases from the text, which helps to quickly understand the theme and core content of the text.

[0071] After completing these preprocessing steps, machine learning or deep learning models can be used to perform more advanced tasks, such as text classification, named entity recognition, question answering systems, etc. These tasks can help better understand and utilize a large amount of text data, thereby providing more accurate services in the auto insurance business.

[0072] In general, text preprocessing is the basis of natural language processing, which directly affects the effect and accuracy of subsequent tasks. Through reasonable preprocessing, the performance of the model can be greatly improved, thus providing users with better services. In future research, more preprocessing methods and technologies can be tried to further improve the efficiency and quality of text processing.

[0073] like Figure 6 As shown, step S220, expanding the optimized short text to obtain preliminary optimized search data, includes:

[0074] S221, establishing a local sensitive hash table;

[0075] S222, using a word vector file to represent the words of the optimized short text as vectors, and obtaining an extended candidate word according to the local sensitive hash table;

[0076] S223: Filter the expanded candidate words to obtain the preliminary optimized search data.

[0077] In the auto insurance business, in order to improve search efficiency and accuracy, the local sensitive hash table (LSH) technology can be used. This is an efficient nearest neighbor search algorithm, especially suitable for similarity search in high-dimensional space. By establishing the local sensitive hash table, large-scale data can be quickly retrieved, greatly improving the speed and efficiency of information processing.

[0078] Then, the words in the optimized short text are represented as vectors using the pre-trained word vector file. These word vectors contain the semantic information of the words, enabling computers to understand and process natural language. Through the local sensitive hash table, other word vectors similar to these word vectors can be quickly found to obtain extended candidate words. This method not only improves the accuracy of the search, but also speeds up data processing.

[0079] Next, the acquired expansion candidates are filtered. This filtering process may be based on a variety of criteria, such as word frequency, word meaning relevance, or specific business needs. The filtered expansion candidates constitute the preliminary optimized search data, which is more accurate and targeted and can better reflect the user's true intention. This step ensures the relevance and practicality of the search results.

[0080] Finally, according to the personalized representation of user search, the preliminary optimized search data is further screened to obtain the final optimized search data. It should be understood that the preliminary optimized search data contains individual data that does not conform to the personalized representation of user search. The data is filtered according to the information of the personalized representation of user search, and the data that conforms to the personalized representation of user search is retained. This process ensures that the search data provided to the seat staff is both relevant and highly personalized, thereby improving work efficiency and customer satisfaction. This method based on local sensitive hash tables not only improves the accuracy of the document topic model, but also enables the seat staff to provide more personalized services based on clearer user intent, thereby significantly improving customer satisfaction and business efficiency.

[0081] like Figure 7 As shown, step 230, generating a personalized representation of user search through the topic vector of the document topic model and the topic distribution of the user search information, includes:

[0082] S231, vectorizing the topics in the document topic model to obtain topic vectors;

[0083] S232, using the document topic model to obtain the topic distribution in the user search information;

[0084] S233: Generate the user search personalized representation using the topic distribution and the topic vector.

[0085] In the auto insurance business, in order to provide more accurate and personalized services, the document topic model must first be vectorized. This means converting the topic into a point in a multidimensional space, where each dimension represents a specific aspect or attribute of the topic. In this way, computers can understand and process complex topic structures, thereby better analyzing and understanding users' search information.

[0086] Next, the document topic model is used to obtain the topic distribution in the user's search information. This process involves text mining and machine learning techniques, such as TF-IDF, LDA and other algorithms. By analyzing the user's search content, the various topics contained in it can be identified and the weight or importance of these topics in the search content can be calculated. In this way, a feature vector about the user's search behavior can be obtained, which integrates the characteristics and preferences of the user's search behavior.

[0087] Then, the obtained user search feature vector is used to generate a personalized representation of the user search. This data structure creates a unique "fingerprint" for each user, which combines the characteristics and preferences of the user's search behavior. This "fingerprint" can help the system better understand the user's needs and provide more accurate and relevant search results.

[0088] In this way, not only the accuracy of the document topic model is improved, but also agents can provide more personalized services based on clearer user intent. The provision of such personalized services significantly improves customer satisfaction and business efficiency. In addition, the application of this method also enables insurance companies to stand out in the fiercely competitive market and provide customers with higher-value products and services.

[0089] In actual operation, this process can be further optimized. For example, deep learning technology can be used to improve the accuracy of topic distribution and the quality of personalized representation. In addition, other data sources, such as user purchase history, geographic location information, etc., can be combined to further improve the user's personalized representation.

[0090] In general, by vectorizing the document topic model, obtaining the topic distribution in the user search information, and generating a personalized representation of the user search, the service quality and efficiency of the auto insurance business can be significantly improved. This method not only helps to improve customer satisfaction, but also helps insurance companies stand out in the fierce market competition. In future research, more optimization methods and application scenarios can be explored to further improve the service level of the auto insurance business.

[0091] S300: Input the optimized search data into an intent recognition model to obtain intent classification.

[0092] These optimized search data are input into the intent recognition model to obtain intent classification. The intent recognition model is a machine learning model that can predict the user's intent category based on the input search data. These categories may include different business needs such as purchasing insurance, claim consultation, policy inquiry, etc.

[0093] like Figure 8 As shown, step S300, inputting the optimized search data into the intent recognition model to obtain intent classification, includes:

[0094] S310, inputting the optimized search data into the intention recognition model, the intention recognition model uses a long short-term memory network as the underlying structure of the intention recognition model, and sets the network depth to 32 and the hidden layer dimension to 256 to obtain an underlying model;

[0095] S320, using the Softmax function to calculate the weight ratio of each hidden layer output in the underlying model;

[0096] S330, performing weighted averaging on the output vectors of all hidden layers according to the weight ratio to obtain a 512-dimensional hidden layer vector;

[0097] S340, inputting the hidden layer vector into the fully connected layer for classification to obtain the intent classification.

[0098] In the auto insurance business, in order to provide more personalized services and improve search efficiency, it is first necessary to obtain the user's search information. This information usually includes keywords, phrases or natural language queries entered by users, which reflect the user's specific needs and intentions. Then, a document topic model is generated based on this user search information. This model can help understand the core topics and key concepts of the user's search content.

[0099] Next, the user search information is preprocessed through the document topic model to obtain optimized search data. This process may include steps such as removing stop words, stemming, and part-of-speech restoration to reduce noise and improve the accuracy of subsequent processing. In addition, word vector technology can be used to convert text into numerical form to facilitate processing by machine learning algorithms.

[0100] Then, these optimized search data are input into the intent recognition model to obtain intent classification. The intent recognition model is a machine learning model that can predict the user's intent category based on the input search data. These categories may include different business needs such as purchasing insurance, claim consultation, policy query, etc. The training of the intent recognition model usually requires a large amount of labeled data, which contains samples of various user intents and corresponding labels. Through supervised learning methods, the model can learn how to extract features from search data and predict intent categories.

[0101] In this process, the Long Short-Term Memory Network (LSTM) is used as the underlying structure of the intent recognition model. The Long Short-Term Memory Network (LSTM) is a special type of recurrent neural network (RNN) that is able to learn long-term dependencies when processing sequence data. This ability makes LSTM perform well in many tasks that require modeling time series or text data. When building a model based on LSTM, some parameters are usually set to adjust the performance and complexity of the model. For example, in this example, the network depth is set to 32, which means that the model will contain 32 LSTM layers. Each layer receives the output of the previous layer as input and produces its own output, so that a deep network structure is formed layer by layer. In addition, the hidden layer dimension is set to 256, which determines the size of the state vector inside each LSTM unit. A larger hidden layer dimension can provide stronger learning and representation capabilities, but it also increases the computational cost of the model and the risk of overfitting. Therefore, choosing the appropriate hidden layer dimension requires a trade-off between model performance and complexity. With such a setting, an underlying model with strong learning and representation capabilities can be constructed, which can capture more information and patterns when processing complex sequence data. However, this is only the first step in model building. Subsequent work such as training, tuning, and evaluation is required to ensure that the model can achieve good results in practical applications.

[0102] In order to extract useful information from the multiple hidden layer outputs of the LSTM model, the Softmax function is used to calculate the weight ratio of each hidden layer output. The Softmax function can map any real number to the interval (0,1), and the sum of its outputs is 1, which makes it possible to interpret the output of the Softmax function as a probability distribution. In this way, the output vectors of all hidden layers can be weighted averaged according to the weight ratio to obtain a 512-dimensional hidden layer vector.

[0103] Among them, the Softmax function is an activation function widely used in deep learning, mainly used for multi-class classification problems. It can convert a real number vector into a probability distribution, in which the value of each element is between 0 and 1, and the sum of all elements is 1. The original intention of the design of the Softmax function is to hope that the influence of the feature on the probability is multiplicative, that is, to amplify larger values ​​and suppress smaller values ​​through exponential operations. This is particularly important in multi-class classification problems because it is necessary to select the most likely category from multiple categories. For example, in the handwritten digit recognition task, Softmax can be used to predict the probability of each digit, so that the digit with the highest probability is selected as the final prediction result. A key feature of the Softmax function is its differentiability, which is very important for neural networks optimized using gradient descent. However, it should be noted that the gradient of Softmax is zero in some areas (such as negative inputs), which may cause neurons to "die", that is, these neurons no longer update weights during training. To avoid this, improved Softmax variants such as "stable_softmax" are sometimes used.

[0104] Finally, this hidden layer vector is input into the fully connected layer for classification to obtain the final intent classification. The fully connected layer is a commonly used neural network structure that can map the input vector to any position in the output space. By training the fully connected layer, a mapping function from the hidden layer vector to the intent category can be learned.

[0105] In this way, not only the accuracy of the document topic model is improved, but also agents can provide more personalized services based on clearer user intent. The provision of such personalized services significantly improves customer satisfaction and business efficiency. In addition, the application of this method also enables insurance companies to stand out in the fiercely competitive market and provide customers with higher-value products and services.

[0106] S400: Generate and display a message prompt according to the intent classification.

[0107] In this embodiment, a message prompt is generated based on the intent classification and displayed to the user. The message prompt is customized based on the user's intent and can provide more accurate and relevant information. For example, if the user's intent is to purchase insurance, the system can display relevant insurance products and price information; if the user's intent is to consult about claims, the system can provide information such as the claims process and contact information. For the agent operation of the insurance company, the message prompt can be given to the agent in real time in the form of a bubble message, so that the agent knows the user's true intention in advance before communicating with the user, thereby improving the agent's operating efficiency.

[0108] It can be seen that in the above scheme, for insurance business, the agent obtains the breakpoint list issued by the system. When the agent obtains the list, the short text information of the user's online search will be obtained synchronously, and then the user's real intention classification is obtained through model training, and then the agent is prompted in real time through bubble messages. In this way, the agent knows the user's real intention in advance before communicating with the user, which improves the agent's work efficiency.

[0109] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0110] In one embodiment, a user intention recognition device based on artificial intelligence is provided, and the intelligent question-answering processing device based on artificial intelligence corresponds to the user intention recognition method based on artificial intelligence in the above embodiment. Fig. 9 As shown, the user intention recognition device includes an acquisition module 101, a processing module 102, an input module 103, and a generation module 104. The detailed description of each functional module is as follows:

[0111] The acquisition module 101 is used to acquire user search information and generate a document topic model according to the user search information;

[0112] The processing module 102 is used to perform text preprocessing on the user search information through the document topic model to obtain optimized search data;

[0113] An input module 103, used for inputting the optimized search data into an intent recognition model to obtain an intent classification;

[0114] The generating module 104 is used to generate and display a message prompt according to the intent classification.

[0115] In one embodiment, the acquisition module 101 is specifically configured to:

[0116] Acquire multiple search topics according to the user search information;

[0117] The search topic that appears most frequently in the user search information is used as the document topic model.

[0118] In one embodiment, the processing module 102 is specifically configured to:

[0119] Performing text segmentation and text denoising processing on the user search information to obtain an optimized short text;

[0120] Expanding the optimized short text to obtain preliminary optimized search data;

[0121] Generate a personalized representation of user search through the topic vector of the document topic model and the topic distribution of the user search information;

[0122] The preliminary optimized search data is filtered and screened according to the user search personalized representation to obtain the optimized search data.

[0123] In one embodiment, the processing module 102 is further configured to:

[0124] Use the Viterbi algorithm to perform text segmentation on the user search information to obtain a preliminary short text;

[0125] Stop words in the preliminary short text are eliminated, and a word vector model is used to perform word vector training on the preliminary short text from which the stop words are eliminated to obtain the optimized short text.

[0126] In one embodiment, the processing module 102 is specifically configured to:

[0127] Build a locality sensitive hash table;

[0128] Using a word vector file, the words of the optimized short text are represented as vectors, and an expansion candidate word is obtained according to the local sensitive hash table;

[0129] The expanded candidate words are filtered to obtain the preliminary optimized search data.

[0130] In one embodiment, the processing module 102 is specifically configured to:

[0131] Vectorizing the topics in the document topic model to obtain topic vectors;

[0132] Using the document topic model to obtain the topic distribution in the user search information;

[0133] The user search personalized representation is generated using the topic distribution and the topic vector.

[0134] In one embodiment, the input module 103 is specifically configured to:

[0135] The optimized search data is input into the intention recognition model, the intention recognition model uses a long short-term memory network as the underlying structure of the intention recognition model, and sets the network depth to 32 and the hidden layer dimension to 256 to obtain an underlying model;

[0136] Use the Softmax function to calculate the weight ratio of each hidden layer output in the underlying model;

[0137] According to the weight ratio, the output vectors of all hidden layers are weighted averaged to obtain a 512-dimensional hidden layer vector;

[0138] The hidden layer vector is input into the fully connected layer for classification to obtain the intent classification.

[0139] The present invention provides a user intention recognition device based on artificial intelligence, which can accurately capture the user's real needs and intentions by analyzing the user's short text search behavior. During the agent's operation, the system can provide personalized prompt information based on these data to help the agent quickly understand customer needs, thereby significantly improving communication efficiency and operation efficiency. This intelligent assistance not only reduces the agent's workload, but also improves customer satisfaction and promotes the improvement of sales conversion rate.

[0140] For the specific limitations of the user intention recognition device based on artificial intelligence, please refer to the limitations of the user intention recognition method above, which will not be repeated here. Each module in the above-mentioned user intention recognition device based on artificial intelligence can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0141] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it realizes the functions or steps of a user intent recognition method based on artificial intelligence on the server side.

[0142] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Fig.11As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it realizes the functions or steps of the client side of a user intention recognition method based on artificial intelligence

[0143] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0144] Obtaining user search information, and generating a document topic model according to the user search information;

[0145] Performing text preprocessing on the user search information through the document topic model to obtain optimized search data;

[0146] Inputting the optimized search data into an intent recognition model to obtain intent classification;

[0147] Generate and display a message prompt based on the intent classification.

[0148] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0149] Obtaining user search information, and generating a document topic model according to the user search information;

[0150] Performing text preprocessing on the user search information through the document topic model to obtain optimized search data;

[0151] Inputting the optimized search data into an intent recognition model to obtain intent classification;

[0152] Generate and display a message prompt based on the intent classification.

[0153] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant descriptions on the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0154] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0155] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0156] The above-described embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the protection scope of the present invention. It should be noted that if non-company software tools or components appear in the embodiments of this application, they are only used for example introduction and do not represent actual use.

Claims

1. A method for identifying user intention based on artificial intelligence, characterized in that: include: Obtaining user search information, and generating a document topic model according to the user search information; Performing text preprocessing on the user search information through the document topic model to obtain optimized search data; Inputting the optimized search data into an intent recognition model to obtain intent classification; Generate and display a message prompt based on the intent classification.

2. The method for identifying user intention based on artificial intelligence as claimed in claim 1, characterized in that: Generating a document topic model according to the user search information includes: Acquire multiple search topics according to the user search information; The search topic that appears most frequently in the user search information is used as the document topic model.

3. The method for identifying user intention based on artificial intelligence as claimed in claim 1, characterized in that: The step of performing text preprocessing on the user search information by using the document topic model to obtain optimized search data includes: Performing text segmentation and text denoising processing on the user search information to obtain an optimized short text; Expanding the optimized short text to obtain preliminary optimized search data; Generate a personalized representation of user search through the topic vector of the document topic model and the topic distribution of the user search information; The preliminary optimized search data is filtered and screened according to the user search personalized representation to obtain the optimized search data.

4. The method for identifying user intention based on artificial intelligence as claimed in claim 3, characterized in that: The performing text segmentation and text denoising processing on the user search information to obtain an optimized short text includes: Use the Viterbi algorithm to perform text segmentation on the user search information to obtain a preliminary short text; Stop words in the preliminary short text are eliminated, and a word vector model is used to perform word vector training on the preliminary short text from which the stop words are eliminated to obtain the optimized short text.

5. The method for identifying user intention based on artificial intelligence as claimed in claim 3, characterized in that: The step of expanding the optimized short text to obtain preliminary optimized search data includes: Build a locality sensitive hash table; Using a word vector file, the words of the optimized short text are represented as vectors, and an expansion candidate word is obtained according to the local sensitive hash table; The expanded candidate words are filtered to obtain the preliminary optimized search data.

6. The method for identifying user intention based on artificial intelligence as claimed in claim 3, characterized in that: The generating of a user search personalized representation by using the topic vector of the document topic model and the topic distribution of the user search information includes: Vectorizing the topics in the document topic model to obtain topic vectors; Using the document topic model to obtain the topic distribution in the user search information; The user search personalized representation is generated using the topic distribution and the topic vector.

7. The method for identifying user intention based on artificial intelligence as claimed in claim 1, characterized in that: The step of inputting the optimized search data into an intent recognition model to obtain an intent classification includes: The optimized search data is input into the intention recognition model, the intention recognition model uses a long short-term memory network as the underlying structure of the intention recognition model, and sets the network depth to 32 and the hidden layer dimension to 256 to obtain an underlying model; Use the Softmax function to calculate the weight ratio of each hidden layer output in the underlying model; According to the weight ratio, the output vectors of all hidden layers are weighted averaged to obtain a 512-dimensional hidden layer vector; The hidden layer vector is input into the fully connected layer for classification to obtain the intent classification.

8. A user intention recognition device based on artificial intelligence, characterized in that: include: An acquisition module, used to acquire user search information and generate a document topic model according to the user search information; A processing module, used for performing text preprocessing on the user search information through the document topic model to obtain optimized search data; An input module, used for inputting the optimized search data into an intent recognition model to obtain an intent classification; A generation module is used to generate and display message prompts according to the intent classification.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the user intention recognition method based on artificial intelligence as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the user intention recognition method based on artificial intelligence as described in any one of claims 1 to 7 are implemented.

Images