Industrial automatic product selection method, device and equipment and storage medium

Abstract

The invention provides an industrial automatic product selection method and device, equipment and a storage medium, and the method comprises the steps: carrying out the intention recognition of a demand description corpus inputted by a user, and obtaining the intention type of a target user; selecting a corresponding cue word according to the intention type, and extracting key information from the demand description corpus based on the cue word; performing encoding processing on the key information to obtain an encoding vector corresponding to the key information, and performing matching retrieval and rearrangement processing on the encoding vector and a preset industrial automation product vector large database to obtain a retrieval result; and inputting the retrieval result and the demand description corpus into the large language model to obtain an industrial selected product reply text corresponding to the demand description corpus, and sending the industrial selected product reply text to the target user. According to the method, recommendation accuracy is optimized through intelligent retrieval and deep learning, user experience is enhanced, efficient intention understanding and search result optimization are provided, real-time data updating is ensured, and industrial automatic product selection efficiency and quality are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to an industrial automation product selection method, device, equipment and storage medium. Background Art

[0002] With the continuous improvement of industrial automation level, various automation products have appeared in the market, including but not limited to sensors, robots, automated assembly lines, etc. When enterprises and individual users choose automation products that suit their needs, they often need to invest a lot of time and energy to analyze and compare the performance parameters, prices, compatibility and other factors of different products.

[0003] At present, the selection of industrial automation products mainly relies on manual operation. Users need to conduct market research according to their own needs, and then make decisions by comparing the technical parameters and application cases of different products. This process is not only inefficient, but also easily limited by personal experience and knowledge level, resulting in suboptimal product selection results. In addition, as the types and quantities of products increase, the complexity and difficulty of manual processing are also increasing.

[0004] Although some software-based solutions have attempted to automatically recommend products through algorithms, these methods often rely on static matching rules, lack sufficient flexibility and intelligence, and cannot fully understand users' specific needs and intentions. Summary of the invention

[0005] The main purpose of the present invention is to solve the technical problem that the existing industrial automation product selection relies on static matching rules and lacks sufficient flexibility and intelligence.

[0006] A first aspect of the present invention provides an industrial automation product selection method, the industrial automation product selection method comprising: Obtaining the demand description corpus input by the target user, and performing intent recognition based on the demand description corpus to obtain the intent type of the target user; Selecting corresponding prompt words according to the intention type, and extracting key information from the demand description corpus based on the prompt words; Encoding the key information to obtain a coding vector corresponding to the key information, and matching, searching and rearranging the coding vector with a preset industrial automation product vector database to obtain a search result; The search results and the demand description corpus are input into a preset large language model to obtain an industrial product selection reply text corresponding to the demand description corpus, and the industrial product selection reply text is sent to the target user.

[0007] Optionally, in a first implementation of the first aspect of the present invention, selecting a corresponding prompt word according to the intent type, and extracting key information from the demand description corpus based on the prompt word includes: Selecting corresponding prompt words according to the intention type, and slicing the demand description corpus to obtain corpus phrases corresponding to the demand description corpus; Using natural language processing technology to perform semantic analysis on the demand description corpus according to the prompt word, and extracting information fragments matching the selected prompt word from the corpus phrases; The information fragments are de-redundanted and structured to obtain key information in the demand description corpus.

[0008] Optionally, in a second implementation of the first aspect of the present invention, the key information is encoded to obtain a coding vector corresponding to the key information, and the coding vector is matched and searched with a preset industrial automation product vector database to obtain a search result including: The key information is converted into a vector form by a preset vector embedding technology to obtain a coding vector of the key information, and the coding vector is matched and retrieved with a preset industrial automation product vector database to obtain k product vectors in the industrial automation product vector database as a matching product vector set; Calculating the priorities of the k product vectors in the matching product vector set according to a preset rearrangement algorithm to obtain the priorities of the product vectors in the matching product vector set, and reordering the k product vectors in the matching product vector set according to the priorities; The product vectors in the reordered matching product vector set are decoded to obtain the retrieval result.

[0009] Optionally, in a third implementation manner of the first aspect of the present invention, the key information is converted into a vector form by a preset vector embedding technology to obtain a coding vector of the key information, and the coding vector is matched and retrieved with a preset industrial automation product vector big database to obtain k product vectors in the industrial automation product vector big database, as a matching product vector set including: Inputting the key information into a preset large language model to generate a hypothetical response text, and inputting the hypothetical response text and the key information into a preset vector embedding model to obtain a corresponding encoding vector; Inputting the encoding vector into a preset industrial automation product vector database, wherein the industrial automation product vector database includes a summary level index and a document level index; Matching the encoding vector with the summary level index in the industrial automation product vector database to determine a corresponding document set; The encoding vector is matched twice with the document level index corresponding to the document set to obtain k product vectors in the industrial automation product vector database as a matching product vector set.

[0010] Optionally, in a fourth implementation of the first aspect of the present invention, before encoding the key information to obtain a coding vector corresponding to the key information, and matching, retrieving and re-arranging the coding vector with a preset industrial automation product vector database to obtain a search result, the method further includes: Collect detailed product data of various industrial automation products to obtain an initial product data set, wherein the detailed product data includes technical parameters, performance indicators and application cases corresponding to the industrial automation products; Performing data cleaning and standardization processing on the initial product data set, and using a preset data labeling technology to label the initial product data set to obtain labeled product data, wherein the labeled product data includes summary data and document data; Input the detailed product data corresponding to the document data into the preset large language model to generate the corresponding hypothetical question text; Inputting the labeled product data and the hypothetical question text into a preset vector embedding model to obtain a product data vector; Clustering the product data vectors corresponding to the summary data to obtain a summary-level index, and constructing a document-level index based on the product data vectors corresponding to the document data and the hypothetical question text; A dynamic update and maintenance mechanism is implemented in the industrial automation product vector big database, and detailed product data of new industrial automation products is introduced regularly to complete the construction of the industrial automation product vector big database.

[0011] Optionally, in a fifth implementation of the first aspect of the present invention, inputting the key information into a preset large language model to generate a hypothetical response text, and inputting the hypothetical response text and the key information into a preset vector embedding model to obtain a corresponding encoding vector includes: The key information and the preset historical storage recall data are input into a preset retrieval evaluator, and the retrieval evaluator performs a relevance evaluation on the key information according to the historical storage recall data to obtain a first relevance score; Determining whether the demand description corpus is high-quality corpus according to a preset scoring threshold and the first relevance score; If not, the key information is input into a preset large language model to generate a hypothetical response text, and the hypothetical response text and the key information are input into a preset vector embedding model to obtain a corresponding encoding vector.

[0012] Optionally, in a sixth implementation manner of the first aspect of the present invention, after performing secondary matching on the encoded vector and the document-level index corresponding to the document set to obtain k product vectors in the industrial automation product vector database as a matching product vector set, the method further includes: Inputting the key information and k product vectors in the industrial automation product vector database into a preset retrieval evaluator, and performing a relevance evaluation based on the k product vectors by the retrieval evaluator to score the relevance of each product vector; Determining whether each product vector meets a preset standard according to a preset scoring threshold and the relevance score; If not, the key information is input into the preset large language model for context expansion to obtain context text, and the context text is re-matched and retrieved with the preset industrial automation product vector big database to obtain k product vectors in the industrial automation product vector big database and update them to the matching product vector set.

[0013] A second aspect of the present invention provides an industrial automation product selection device, the industrial automation product selection device comprising: An intention recognition module is used to obtain the demand description corpus input by the target user, and perform intention recognition based on the demand description corpus to obtain the intention type of the target user; An information extraction module, used to select corresponding prompt words according to the intention type, and extract key information from the demand description corpus based on the prompt words; A retrieval module is used to encode the key information to obtain a coding vector corresponding to the key information, and match the coding vector with a preset industrial automation product vector database for retrieval and rearrangement to obtain a retrieval result; A reply module is used to input the search results and the demand description corpus into a preset large language model, obtain the industrial product selection reply text corresponding to the demand description corpus, and send the industrial product selection reply text to the target user.

[0014] The third aspect of the present invention provides an industrial automation product selection device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the industrial automation product selection equipment executes the steps of the above-mentioned industrial automation product selection method.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium, which stores instructions that, when executed on a computer, enable the computer to execute the steps of the above-mentioned industrial automation product selection method.

[0016] The above industrial automation product selection method, device, equipment and storage medium obtain the target user's intention type by performing intent recognition on the demand description corpus input by the user; select the corresponding prompt word according to the intention type, and extract key information from the demand description corpus based on the prompt word; encode the key information to obtain the encoding vector corresponding to the key information, and match the encoding vector with the preset industrial automation product vector database for retrieval and rearrangement to obtain the retrieval result; input the retrieval result and the demand description corpus into the large language model to obtain the industrial product selection reply text corresponding to the demand description corpus, and send it to the target user. This method optimizes the recommendation accuracy through intelligent retrieval and deep learning, enhances the user experience, provides efficient intent understanding and search result optimization, ensures real-time data update, and significantly improves the efficiency and quality of industrial automation product selection.

[0017] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic diagram of an embodiment of an industrial automation product selection method in an embodiment of the present invention; Figure 2 A schematic diagram of an embodiment of an industrial automation product selection device in an embodiment of the present invention; Figure 3 It is a schematic diagram of an embodiment of industrial automation product selection equipment in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. 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.

[0021] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0022] To facilitate understanding of this embodiment, firstly, an industrial automation product selection method disclosed in an embodiment of the present invention is described in detail. Figure 1 As shown, the method comprises the following steps: 101. Obtain the demand description corpus input by the target user, and perform intent recognition based on the demand description corpus to obtain the target user's intent type; In one embodiment of the present invention, the system performs in-depth intent recognition on the demand description corpus input by the target user to accurately obtain the user's intent type and provide the most appropriate product recommendation accordingly. In this process, the system first collects the user's input, which can be a detailed description of a specific industrial automation demand, such as "I need a robot arm that can work stably in a high temperature environment for precision assembly of electronic devices." Once these demand description corpora are obtained, the system uses natural language processing technology to analyze the corpus. The key to this step is to use deep learning models, such as transformer-based models, to process and parse text data. These models have been trained with a large amount of data and can understand the complex semantics and structure of human language. The system uses these models to extract key features of the input text, such as keywords, phrases and their contextual associations. After extracting the key features, the system further determines the user's intent type through an intent recognition algorithm. This algorithm is based on a concept called "intention vector space model", which maps the user's demand description to a predefined intent space. In the specific implementation steps, the system first processes the user's input corpus through word segmentation and annotation, then uses the embedding layer to convert the text into vectors, and then processes these vectors through a multi-layer transformer network, and finally outputs the predicted intent vector. For example, if the user's input is about a robot arm that needs to work in a high-temperature environment, the system can identify "high-temperature stability" and "precision assembly" as key intents and match them to the corresponding robot product category.

[0023] 102. Select corresponding prompt words according to the intent type, and extract key information from the demand description corpus based on the prompt words; In one embodiment of the present invention, the selecting corresponding prompt words according to the intent type and extracting key information from the demand description corpus based on the prompt words includes: selecting corresponding prompt words according to the intent type, and slicing the demand description corpus to obtain corpus phrases corresponding to the demand description corpus; using natural language processing technology to semantically parse the demand description corpus according to the prompt words, and extracting information fragments matching the selected prompt words from the corpus phrases; performing de-redundancy processing and structuring processing on the information fragments to obtain key information in the demand description corpus.

[0024] Specifically, after the intent type of the user input is successfully identified through natural language processing technology, the prompt word matching this intent is selected. The selection of prompt words is based on a pre-defined mapping table of intent to prompt words, in which each intent type is associated with a set of keywords or phrases that best reflect the characteristics of the intent. For example, if the intent type identified by the system is "high temperature operation", the relevant prompt words include "heat resistance", "high temperature stability", etc. Next, the system slices the user's demand description corpus, which involves word segmentation and syntactic analysis of the text, aiming to decompose continuous text data into smaller pieces that are easier to handle, namely corpus phrases. These phrases will be organized according to syntactic structure and logical relationship, so as to carry out subsequent semantic analysis more accurately. Next, the system uses natural language processing technology to perform semantic analysis on the above corpus phrases, paying special attention to the information fragments that match the previously selected prompt words. In this step, each corpus phrase will be evaluated for its relevance to the prompt word. Specifically, the system uses methods such as TF-IDF, in which the system can identify phrases with high weights in the demand description that are related to the selected prompt words, thereby more accurately capturing the user's specific needs. Finally, the information fragments extracted from the phrases will be de-redundant and structured, with the aim of eliminating repetitive or irrelevant information and logically reorganizing and optimizing the remaining content. This process ensures that the key information finally extracted is refined and highly relevant, and can be directly used for subsequent product matching and recommendations. For example, if the user requests "find an automated valve that can work stably in a variety of chemical environments", the system will identify and extract keywords such as "chemical stability" and "multi-environment adaptability" through the above steps to ensure that the recommended products meet all listed working environment requirements.

[0025] 103. Encode the key information to obtain a coding vector corresponding to the key information, and match, retrieve and rearrange the coding vector with a preset industrial automation product vector database to obtain a search result; In one embodiment of the present invention, the key information is encoded to obtain a coding vector corresponding to the key information, and the coding vector is matched and retrieved with a preset industrial automation product vector database to obtain the search result, including: converting the key information into a vector form through a preset vector embedding technology to obtain the coding vector of the key information, and matching and searching the coding vector with a preset industrial automation product vector database to obtain k product vectors in the industrial automation product vector database as a matching product vector set; calculating the priority of the k product vectors in the matching product vector set according to a preset rearrangement algorithm to obtain the priority of each product vector in the matching product vector set, and reordering the k product vectors in the matching product vector set according to the priority; decoding the product vectors in the reordered matching product vector set to obtain the search result.

[0026] Specifically, the system will perform vector embedding processing on the key information extracted from the user's demand description. The core of this processing step is to convert natural language text into numerical vectors that can be processed by computer algorithms, which is usually achieved through a pre-trained embedding model such as Word2Vec or BERT. These models can capture the semantic features of words and map them into a high-dimensional space, where semantically similar words are mapped to similar points. These vectors not only contain the basic meaning of the words, but also contain the relative relationship with other words and contextual information, making the vector representation of the entire demand description highly informative and discriminative. After obtaining the encoding vector of the key information, the system further matches and retrieves these vectors with a preset large database of industrial automation product vectors. This database has pre-constructed vector representations of various industrial automation products, and each product vector is a high-dimensional numerical expression of the product's attributes and functions. By calculating the similarity between the user's demand vector and each product vector in the database, the system can quickly locate the most matching product. Subsequently, the system sorts the matched product vector set according to the calculated similarity score, and recommends those with the highest scores first. This step involves complex sorting algorithms, including machine learning algorithms to optimize the accuracy and response speed of sorting. After sorting, the system decodes these vectors and converts them back into product descriptions and recommendation information that are easy for users to understand.

[0027] Furthermore, the key information is converted into a vector form through a preset vector embedding technology to obtain a coding vector of the key information, and the coding vector is matched and retrieved with a preset industrial automation product vector big database to obtain k product vectors in the industrial automation product vector big database as a matching product vector set, including: inputting the key information into a preset large language model to generate a hypothetical response text, and inputting the hypothetical response text and the key information into a preset vector embedding model to obtain a corresponding coding vector; inputting the coding vector into a preset industrial automation product vector big database, wherein the industrial automation product vector big database includes a summary level index and a document level index; matching the coding vector with the summary level index in the industrial automation product vector big database once to determine the corresponding document set; matching the coding vector with the document level index corresponding to the document set twice to obtain k product vectors in the industrial automation product vector big database as a matching product vector set.

[0028] Specifically, after receiving the key information from the user, the information will first be input into a preset large language model. This large language model is built based on advanced language processing technologies such as BERT or GPT, which can understand and generate hypothetical response texts that are consistent with the semantics of the input information. For example, if the user inputs "need a valve suitable for high temperature and high pressure environment", the large language model generates hypothetical responses including "looking for a valve that can withstand 100 degrees Celsius and 10MPa pressure". The purpose of this step is to capture the deep characteristics of user needs by simulating a richer semantic environment. Next, these generated hypothetical response texts are input into the preset vector embedding model together with the original key information. After that, these encoded vectors are input into the preset industrial automation product vector database. This database not only includes vector representations of various products, but also sets summary-level and document-level indexes for efficient retrieval and matching. The first step is to match with the summary-level index. This step is mainly to quickly screen out roughly matching product categories and reduce the burden of subsequent processing. These initially screened vectors are then matched against the document-level index for a more detailed secondary match, which is based on richer product descriptions and detailed information to ensure that the final product vector set is highly relevant to user needs. For example, the system identifies all product vectors marked as "high-temperature and high-pressure valves" during the initial screening, and further confirms whether these products can meet the specific use conditions of 100 degrees Celsius and 10MPa during the secondary screening.

[0029] Furthermore, before encoding the key information to obtain the encoding vector corresponding to the key information, matching, retrieving and rearranging the encoding vector with a preset industrial automation product vector database to obtain the search result, it also includes: collecting detailed product data of various industrial automation products to obtain an initial product data set, wherein the detailed product data includes technical parameters, performance indicators and application cases corresponding to the industrial automation products; performing data cleaning and standardization on the initial product data set, and using a preset data labeling technology to label the initial product data set to obtain labeled product data, wherein the labeled product The data includes summary data and document data; the detailed product data corresponding to the document data is input into a preset large language model to generate corresponding hypothetical question text; the labeled product data and the hypothetical question text are input into a preset vector embedding model to obtain a product data vector; the product data vector corresponding to the summary data is clustered to obtain a summary level index, and a document level index is constructed based on the product data vector corresponding to the document data and the hypothetical question text; a dynamic update and maintenance mechanism is implemented in the industrial automation product vector big database, and detailed product data of new industrial automation products is regularly introduced to complete the construction of the industrial automation product vector big database.

[0030] Specifically, first, the system will collect detailed data of various industrial automation products from multiple data sources, including but not limited to technical parameters, performance indicators and application cases. For example, for an automated assembly line, the system will collect its technical parameters such as maximum load, speed, accuracy, and performance cases in different application scenarios such as automobile manufacturing and electronic assembly. These original data sets are the basis for all subsequent data processing and analysis. Next, the collected initial product data sets are cleaned and standardized. Data cleaning includes steps such as removing duplicate records, filling missing values, and correcting erroneous data, while standardization ensures that the data format is unified for subsequent operations. For example, the speed parameters of all products are converted to the same unit (such as meters per second) to ensure data consistency and comparability. Subsequently, the system will further process the cleaned data using a preset data labeling technology. Labeling involves assigning one or more labels to each data point based on the product's function, application scenario, or any other key feature. For example, a sensor with high precision and fast response characteristics will be labeled with "high precision" and "fast". Labeled data not only helps to retrieve and match products faster, but also enhances the semantic expression ability of data. In addition, the labeled product data and the hypothetical question text generated from the data are input into the preset vector embedding model. The purpose of vectorization is to convert the text and labels into a form that can be directly processed by the machine learning algorithm, which is crucial for subsequent data analysis and product matching. Next, these product data vectors are clustered to build a summary-level index, and a document-level index is built based on the vectors generated from the detailed document data and question text. Finally, the system implements a dynamic update and maintenance mechanism to regularly introduce new industrial automation product data, update and expand the database. This ensures that the database is always up to date and can reflect the latest product and technology changes on the market.

[0031] Furthermore, the step of inputting the key information into a preset large language model to generate a hypothetical response text, and inputting the hypothetical response text and the key information into a preset vector embedding model to obtain a corresponding encoding vector includes: inputting the key information and preset historical storage recall data into a preset retrieval evaluator, and performing a relevance evaluation on the key information according to the historical storage recall data by the retrieval evaluator to obtain a first relevance score; judging whether the demand description corpus is a high-quality corpus according to a preset score threshold and the first relevance score; if not, inputting the key information into a preset large language model to generate a hypothetical response text, and inputting the hypothetical response text and the key information into a preset vector embedding model to obtain a corresponding encoding vector.

[0032] Specifically, the system inputs the user's key information together with the preset historical storage recall data into the retrieval evaluator. This retrieval evaluator is built based on machine learning technology, especially using models such as support vector machines (SVM) or deep neural networks that can process high-dimensional data and make classification decisions. The main task of the retrieval evaluator is to evaluate the relevance of the input data, that is, to determine the degree of match between the user's key information and similar cases in the historical data. After obtaining the first relevance score, the system compares this score with the preset score threshold. If the score is higher than the threshold, it means that the submitted demand description corpus is of high quality, that is, it is highly relevant to the existing data and can be directly used for subsequent processing. If the score is lower than the threshold, it means that the quality of the demand description corpus is insufficient and further processing is required to improve its quality. In this case, the system will re-input the key information into a preset large language model, such as a model based on GPT-3, which can generate corresponding hypothetical response text based on the input text. This step is to enrich the context and details of the original data and enhance the richness and diversity of its expression. The generated hypothetical response text is then input into the vector embedding model together with the original key information.

[0033] Furthermore, after performing secondary matching on the encoding vector and the document-level index corresponding to the document set to obtain k product vectors in the industrial automation product vector big database as a matching product vector set, it also includes: inputting the key information and the k product vectors in the industrial automation product vector big database into a preset retrieval evaluator, performing relevance evaluation based on the k product vectors by the retrieval evaluator, and scoring the relevance of each product vector; judging whether each product vector meets the preset standard based on a preset scoring threshold and the relevance score; if not, inputting the key information into a preset large language model for context expansion to obtain context text, and re-matching and retrieving the context text with the preset industrial automation product vector big database to re-obtain k product vectors in the industrial automation product vector big database, and updating them as the matching product vector set.

[0034] Specifically, first, the system inputs the key information provided by the user and the k product vectors in the preset industrial automation product vector database into the retrieval evaluator. This retrieval evaluator is based on deep learning technology and can process and analyze high-dimensional data to calculate the correlation between each product vector and the user's key information. After obtaining the relevance score, the system compares these scores with the preset score threshold. If the relevance score of the product vector exceeds the threshold, it indicates that the product is highly relevant to the user's needs and can be recommended to the user. If the score is lower than the threshold, it means that there is a certain gap between the current key information and the product vector, and further processing is required to improve the quality of matching. In this case, the system re-enters the user's key information into the preset large language model for context expansion. This large language model, such as GPT-3 or BERT, uses its powerful language understanding ability to generate relevant context text based on the input key information. These context texts not only enrich the details of the original information, but also add additional semantic levels, making the re-matching process more accurate. The generated context text is then re-entered into the industrial automation product vector database together with the original key information for matching retrieval. This retrieval not only considers the original product vectors, but also introduces a new scoring mechanism and matching algorithm to ensure that the user needs after the context expansion can be better captured. For example, if the key information initially provided by the user is "automatic robotic arms for high-temperature operation", and the initial evaluation finds that there are insufficient matching products, the system generates relevant contextual information including "suitable for precision operations in high-temperature environments" through a large language model. This information is used for matching again, and more product options that were not matched before but are actually very suitable will be identified.

[0035] 104. Input the search results and the demand description corpus into a preset large language model to obtain the industrial product selection response text corresponding to the demand description corpus, and send the industrial product selection response text to the target user.

[0036] In one embodiment of the present invention, the system needs to integrate a preset large language model, such as a model based on GPT-3 or BERT, which has been widely used in various natural language processing tasks in the past few years, including text generation, question-answering systems, and semantic understanding. In our application scenario, the main task of the model is to generate a coherent and accurate reply text based on the input demand description corpus and retrieval results. This generation is based on a process called conditional language generation, in which the model not only responds to the input text, but also takes into account contextual information to generate highly relevant and informative answers.

[0037] Specifically, the data input into the large language model includes the user's demand description and the product data retrieved by the system. The demand description provides the user's specific needs and expectations, while the product data includes the details of the best matching products found by the system based on these needs. After the model receives the input, a complex generation process is used to construct the reply text. This process involves a multi-layer neural network in the decoder part, which can generate text word by word (or word by word), and each generation is based on the previous output and the overall context information. The generated text is gradually constructed until a complete reply is formed. The quality and relevance of the text generation are directly affected by the quality of the input data and the completeness of the model training. For example, if a user queries "Looking for a high corrosion resistance pump suitable for a chemical plant", the system finds several pumps with high corrosion resistance through the retrieval evaluator and inputs this information into the large language model. The reply generated by the model may be: "Based on your needs, we recommend the following high corrosion resistance pumps: [Pump model 1], suitable for chemical media; [Pump model 2], providing additional high temperature resistance; etc." This reply not only meets the user's specific technical needs, but also provides multiple options for the user to choose from.

[0038] Specifically, in this embodiment, the language recognition function and the conversation history storage and recall function are specially introduced. These two functions jointly improve the user's interactive experience and the response efficiency of the system. First, as an important part of the system, the language recognition function uses a pre-trained language model to identify the language tendency of the user input, thereby providing accurate multi-language support. The core of this function is to determine the language that the user may use, and on this basis adjust the output of the reply generation process to ensure the natural fluency of communication. For example, when the user uses a mixed input of English and Chinese, such as mentioning professional terms such as "PLC (Programmable Logic Controller)", the language recognition function can identify and process this language mixture to ensure that the corresponding terms in the reply are correctly displayed, whether in English or Chinese. In addition, the language recognition function can also provide appropriate prompt words when the system generates a reply, which not only enhances the relevance and accuracy of the reply, but also enables the intelligent assistant to flexibly switch between Chinese and English, greatly expanding the user group of the system. The implementation of this function relies on advanced machine learning technology and deep neural networks to ensure high-efficiency and high-accuracy language recognition. Next, the conversation history storage and recall function provides effective support for the performance of the large language model. This feature continuously tracks and stores the historical content of the conversation during the interaction between the user and the system. Whenever the user initiates a new input, this feature will recall the previous conversation history from the cache and provide this information to intent recognition and other related functions. This mechanism significantly improves the system's ability to handle multiple rounds of conversations, so that each user's query can be understood and responded to in the context of previous exchanges, thereby improving the coherence of the conversation and the overall intelligence of the system. For example, when a user asks a question about the equipment specifications again after querying a specific type of industrial automation equipment, the conversation history storage recall function can link the previous query content with the current needs to ensure the consistency and accuracy of the response. This not only reduces the need for users to repeatedly enter the same information, but also makes the conversation more efficient and user-friendly.

[0039] In this embodiment, the intent type of the target user is obtained by performing intent recognition on the demand description corpus input by the user; the corresponding prompt word is selected according to the intent type, and key information is extracted from the demand description corpus based on the prompt word; the key information is encoded to obtain the encoding vector corresponding to the key information, and the encoding vector is matched, retrieved and rearranged with the preset industrial automation product vector database to obtain the retrieval result; the retrieval result and the demand description corpus are input into the large language model to obtain the industrial product selection reply text corresponding to the demand description corpus, and sent to the target user. This method optimizes the recommendation accuracy through intelligent retrieval and deep learning, enhances the user experience, provides efficient intent understanding and search result optimization, ensures real-time data update, and significantly improves the efficiency and quality of industrial automation product selection.

[0040] The above describes the industrial automation selection method in the embodiment of the present invention. The following describes the industrial automation selection device in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, an industrial automation selection device includes: The intention recognition module 201 is used to obtain the demand description corpus input by the target user, and perform intention recognition based on the demand description corpus to obtain the intention type of the target user; An information extraction module 202, configured to select a corresponding prompt word according to the intention type, and extract key information from the demand description corpus based on the prompt word; The retrieval module 203 is used to encode the key information to obtain the encoding vector corresponding to the key information, and match the encoding vector with a preset industrial automation product vector database for retrieval and rearrangement to obtain a retrieval result; The reply module 204 is used to input the search results and the demand description corpus into a preset large language model, obtain the industrial product selection reply text corresponding to the demand description corpus, and send the industrial product selection reply text to the target user.

[0041] In an embodiment of the present invention, the industrial automation product selection device runs the above industrial automation product selection method, and the industrial automation product selection device obtains the target user's intention type by performing intent recognition on the demand description corpus input by the user; selects the corresponding prompt word according to the intention type, and extracts key information from the demand description corpus based on the prompt word; encodes the key information to obtain the encoding vector corresponding to the key information, and matches the encoding vector with the preset industrial automation product vector database for retrieval and rearrangement to obtain the retrieval result; inputs the retrieval result and the demand description corpus into the large language model to obtain the industrial product selection reply text corresponding to the demand description corpus, and sends it to the target user. This method optimizes the recommendation accuracy through intelligent retrieval and deep learning, enhances the user experience, provides efficient intent understanding and search result optimization, ensures real-time data updates, and significantly improves the efficiency and quality of industrial automation product selection.

[0042] above Figure 2 The industrial automation product selection device in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The industrial automation product selection device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0043] Figure 3It is a structural diagram of an industrial automation selection device provided by an embodiment of the present invention. The industrial automation selection device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the industrial automation selection device 300. Furthermore, the processor 310 can be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the industrial automation selection device 300 to implement the steps of the above-mentioned industrial automation selection method.

[0044] The industrial automation selection device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that Figure 3 The structure of the industrial automation product selection equipment shown does not constitute a limitation on the industrial automation product selection equipment provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0045] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the industrial automation product selection method.

[0046] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0047] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0048] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, 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 embodiments of the present invention.

Claims

1. An industrial automation product selection method, characterized in that: The industrial automation product selection method comprises: Obtaining the demand description corpus input by the target user, and performing intent recognition based on the demand description corpus to obtain the intent type of the target user; Selecting corresponding prompt words according to the intention type, and extracting key information from the demand description corpus based on the prompt words; Encoding the key information to obtain a coding vector corresponding to the key information, and matching, searching and rearranging the coding vector with a preset industrial automation product vector database to obtain a search result; The search results and the demand description corpus are input into a preset large language model to obtain an industrial product selection reply text corresponding to the demand description corpus, and the industrial product selection reply text is sent to the target user.

2. The industrial automation product selection method according to claim 1, characterized in that: The selecting a corresponding prompt word according to the intention type, and extracting key information from the demand description corpus based on the prompt word includes: Selecting corresponding prompt words according to the intention type, and slicing the demand description corpus to obtain corpus phrases corresponding to the demand description corpus; Using natural language processing technology to perform semantic analysis on the demand description corpus according to the prompt word, and extracting information fragments matching the selected prompt word from the corpus phrases; The information fragments are de-redundanted and structured to obtain key information in the demand description corpus.

3. The industrial automation product selection method according to claim 1, characterized in that: The encoding process is performed on the key information to obtain a coding vector corresponding to the key information, and the coding vector is matched and searched with a preset industrial automation product vector database to obtain a search result including: The key information is converted into a vector form by a preset vector embedding technology to obtain a coding vector of the key information, and the coding vector is matched and retrieved with a preset industrial automation product vector database to obtain k product vectors in the industrial automation product vector database as a matching product vector set; Calculating the priorities of the k product vectors in the matching product vector set according to a preset rearrangement algorithm to obtain the priorities of the product vectors in the matching product vector set, and reordering the k product vectors in the matching product vector set according to the priorities; The product vectors in the reordered matching product vector set are decoded to obtain the retrieval result.

4. The industrial automation product selection method according to claim 3, characterized in that: The key information is converted into a vector form by a preset vector embedding technology to obtain a coding vector of the key information, and the coding vector is matched and retrieved with a preset industrial automation product vector database to obtain k product vectors in the industrial automation product vector database as a matching product vector set including: Inputting the key information into a preset large language model to generate a hypothetical response text, and inputting the hypothetical response text and the key information into a preset vector embedding model to obtain a corresponding encoding vector; Inputting the encoding vector into a preset industrial automation product vector database, wherein the industrial automation product vector database includes a summary level index and a document level index; Matching the encoding vector with the summary level index in the industrial automation product vector database to determine a corresponding document set; The encoding vector is matched twice with the document level index corresponding to the document set to obtain k product vectors in the industrial automation product vector database as a matching product vector set.

5. The industrial automation product selection method according to claim 4, characterized in that: Before encoding the key information to obtain a coding vector corresponding to the key information, matching, searching and rearranging the coding vector with a preset industrial automation product vector database to obtain a search result, the method further includes: Collect detailed product data of various industrial automation products to obtain an initial product data set, wherein the detailed product data includes technical parameters, performance indicators and application cases corresponding to the industrial automation products; Performing data cleaning and standardization processing on the initial product data set, and using a preset data labeling technology to label the initial product data set to obtain labeled product data, wherein the labeled product data includes summary data and document data; Input the detailed product data corresponding to the document data into the preset large language model to generate the corresponding hypothetical question text; Inputting the labeled product data and the hypothetical question text into a preset vector embedding model to obtain a product data vector; Clustering the product data vectors corresponding to the summary data to obtain a summary-level index, and constructing a document-level index based on the product data vectors corresponding to the document data and the hypothetical question text; A dynamic update and maintenance mechanism is implemented in the industrial automation product vector big database, and detailed product data of new industrial automation products is introduced regularly to complete the construction of the industrial automation product vector big database.

6. The industrial automation product selection method according to claim 5, characterized in that: The step of inputting the key information into a preset large language model to generate a hypothetical response text, and inputting the hypothetical response text and the key information into a preset vector embedding model to obtain a corresponding encoding vector includes: The key information and the preset historical storage recall data are input into a preset retrieval evaluator, and the retrieval evaluator performs a relevance evaluation on the key information according to the historical storage recall data to obtain a first relevance score; Determining whether the demand description corpus is high-quality corpus according to a preset scoring threshold and the first relevance score; If not, the key information is input into a preset large language model to generate a hypothetical response text, and the hypothetical response text and the key information are input into a preset vector embedding model to obtain a corresponding encoding vector.

7. The industrial automation product selection method according to claim 6, characterized in that: After performing secondary matching between the encoding vector and the document level index corresponding to the document set to obtain k product vectors in the industrial automation product vector database as a matching product vector set, the method further includes: Inputting the key information and k product vectors in the industrial automation product vector database into a preset retrieval evaluator, and performing a relevance evaluation based on the k product vectors by the retrieval evaluator to score the relevance of each product vector; Determining whether each product vector meets a preset standard according to a preset scoring threshold and the relevance score; If not, the key information is input into the preset large language model for context expansion to obtain context text, and the context text is re-matched and retrieved with the preset industrial automation product vector big database to obtain k product vectors in the industrial automation product vector big database and update them to the matching product vector set.

8. An industrial automation product selection device, characterized in that: The industrial automation product selection device comprises: An intention recognition module is used to obtain the demand description corpus input by the target user, and perform intention recognition based on the demand description corpus to obtain the intention type of the target user; An information extraction module, used to select corresponding prompt words according to the intention type, and extract key information from the demand description corpus based on the prompt words; A retrieval module is used to encode the key information to obtain a coding vector corresponding to the key information, and match the coding vector with a preset industrial automation product vector database for retrieval and rearrangement to obtain a retrieval result; A reply module is used to input the search results and the demand description corpus into a preset large language model, obtain the industrial product selection reply text corresponding to the demand description corpus, and send the industrial product selection reply text to the target user.

9. An industrial automation selection device, characterized in that: The industrial automation selection device comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory so that the industrial automation selection device executes the steps of the industrial automation selection method as described in any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the industrial automation product selection method as described in any one of claims 1-7 are implemented.

Images