Building material retrieval method

Abstract

The invention provides a building material retrieval method, which comprises the following steps of: acquiring a building material retrieval request described by a natural language of a user, performing deep semantic analysis by using a large language model to extract key information such as names, specifications, performances, scenes and the like, and generating a keyword set; performing synonym, hypernym, hypernym, industry term and associated word expansion on the keywords based on material attributes and semantic correlation to form an expanded keyword set; comprehensively retrieving multi-dimensional attributes such as names, specifications, performance, manufacturers, prices and the like in the building material database to obtain preliminary results; and performing weighted scoring and sorting by using multiple factors such as a semantic matching degree, an attribute matching degree, user evaluation and a market trend, and outputting a visual chart. According to the method, the building material retrieval accuracy, efficiency and user experience can be improved, and diversified retrieval requirements of professional users in the building industry are met.

Description

Technical Field

[0001] This invention relates to the field of building material information retrieval technology, and more specifically, to a building material retrieval method. Background Technology

[0002] In the field of building materials retrieval, with the rapid development of the construction industry, the requirements for the types, specifications, and performance of building materials are becoming increasingly diversified. Traditional building materials retrieval methods mainly rely on keyword matching and simple classification searches, requiring users to accurately input information such as the name and specifications of the building materials to obtain relevant results. However, this method has many limitations. First, users may not be able to accurately describe the detailed information of the required building materials, leading to inaccurate or incomplete search results. Second, traditional retrieval methods cannot understand the semantic information in the natural language descriptions entered by users, making it difficult to capture the multi-dimensional characteristics of building materials, such as performance indicators and application scenarios. In addition, traditional methods perform poorly when handling fuzzy queries and multi-condition comprehensive searches, failing to meet users' high requirements for the quality and efficiency of search results.

[0003] With the development of big data and artificial intelligence technologies, natural language processing and large language models have been widely applied in the field of information retrieval. Large language models, through deep learning technology, can perform deep semantic understanding and generation of natural language text, providing new ideas for overcoming the limitations of traditional retrieval methods. However, in the field of building materials retrieval, how to effectively utilize the semantic parsing and generation capabilities of large language models to achieve multi-dimensional retrieval and accurate matching of building materials remains a problem to be solved. Most existing retrieval systems based on large language models focus on general domains, lacking in-depth understanding and processing capabilities for the specific attributes and semantics of the building materials industry, and thus failing to meet the needs of professional users in the construction industry.

[0004] In implementing the embodiments of the present invention, the prior art has at least the following problems or defects: First, traditional retrieval methods cannot effectively process semantic information in natural language descriptions, resulting in inaccurate retrieval results; second, existing retrieval systems based on large language models lack the ability to process the unique attributes of the building materials industry, and cannot achieve multi-dimensional retrieval and accurate matching; third, existing systems are inefficient in processing fuzzy queries and multi-condition comprehensive searches, and cannot meet users' requirements for the quality and efficiency of retrieval results. Summary of the Invention

[0005] This invention provides a method for retrieving building materials, comprising: Obtain building material retrieval request information, which includes a natural language description input by the user; The pre-trained large language model is used to perform deep semantic parsing on the building material retrieval request information, extract key information from the retrieval request, and generate a set of retrieval keywords. The key information includes the name, specifications, performance indicators, and application scenarios of the building materials. Based on the property characteristics and semantic relevance of building materials, the set of search keywords is expanded in multiple dimensions to generate an expanded set of keywords. The multi-dimensional expansion includes synonym replacement, hyponym and hypernym expansion, industry terminology matching, and semantic association word generation. In the building materials database, a multi-dimensional search is performed based on the extended keyword set. The multi-dimensional search includes a comprehensive search of the building materials' name, specifications, performance indicators, manufacturers, and price attributes to obtain preliminary search results. The preliminary search results are subjected to a multi-factor relevance score, which includes semantic matching score, attribute matching score, user evaluation score and market trend score. The preliminary search results are sorted according to the comprehensive relevance score to obtain the final search results. The final search results are output to the user interface in the form of visual charts, including bar charts, line charts, and pie charts.

[0006] Furthermore, the process of using a pre-trained large language model to perform deep semantic parsing on the building material retrieval request information, extracting key information from the retrieval request, and generating a set of retrieval keywords includes: A pre-trained large language model is used to perform deep semantic understanding on the building material retrieval request information, and key information is extracted as a set of retrieval keywords. The set of search keywords is normalized to remove duplicate and redundant keywords; The set of search keywords is semantically enhanced by generating semantic feature vectors related to the search keywords through a large language model.

[0007] Furthermore, the multi-dimensional expansion of the search keyword set based on building material attribute characteristics and semantic relevance to generate an expanded keyword set includes: Based on the characteristics of building material properties, synonym replacement and hyponym expansion are performed on each search keyword in the search keyword set; Using semantic relevance analysis of a large language model, related extended keywords are generated for each search keyword; The extended keyword set was expanded by combining it with a professional terminology dictionary for the building materials industry; The extended keyword set is deduplicated and filtered.

[0008] Furthermore, in the building materials database, a multi-dimensional search is performed based on the extended keyword set to obtain preliminary search results, including: The building material entries in the building materials database are indexed in multiple dimensions, including name index, specification index, performance index, manufacturer index, and price index. Based on the extended keyword set, searches are performed in the indexes of each dimension to obtain multi-dimensional search results; The search results from multiple dimensions are merged and deduplicated to obtain preliminary search results; The initial search results are initially filtered to remove building material entries that do not meet the search requirements.

[0009] Furthermore, the step of performing multi-factor relevance scoring on the preliminary search results to obtain the final search results includes: For each building material item in the preliminary search results, a semantic matching score is calculated. Based on a large language model, the semantic similarity between the search request and the description of the building material item is calculated. The formula for calculating the semantic matching score is as follows:

[0010] in, Indicates the semantic matching score. This represents the semantic feature vector of the retrieval request. The semantic feature vector representing the building material item. For adjustment coefficients; Each building material item in the preliminary search results is scored for attribute matching degree. Each building material item in the preliminary search results will be evaluated and rated by users. Each building material item in the preliminary search results will be scored for market trends. The semantic matching score, attribute matching score, user evaluation score, and market trend score are weighted and summed to obtain the comprehensive relevance score for each building material item. The preliminary search results are sorted based on the comprehensive relevance score, and building material entries with scores higher than a preset threshold are selected as the final search results.

[0011] Furthermore, outputting the final search results to the user interface in the form of a visual chart includes: Provide a detailed information page for the search results, which includes a detailed description of the building materials, manufacturer information, and user reviews; The user interface provides interactive features that allow users to further filter and sort search results based on attributes; Generate dynamic, interactive visualizations; The system provides a function to export search results, which supports multiple formats.

[0012] Furthermore, the building material entries in the building materials database include the name, specifications, performance indicators, manufacturer, and price of the building materials; The building materials database is updated regularly. The building materials database supports dynamic expansion; The data in the building materials database undergoes quality control. The building materials database adopts a distributed storage architecture; The building materials database integrates a data mining module; The building materials database supports multilingual retrieval.

[0013] Furthermore, the step of expanding the search keyword set in multiple dimensions based on building material attribute characteristics and semantic relevance to generate an expanded keyword set also includes: By leveraging the context-aware capabilities of large language models, context-related extended keywords are generated for each search keyword in the search keyword set; Perform semantic validation on each extended keyword in the extended keyword set; The expanded keyword set is dynamically optimized based on user feedback and historical search data.

[0014] Furthermore, the step of performing multi-factor relevance scoring on the preliminary search results to obtain the final search results also includes: Each building material entry in the preliminary search results is preprocessed with text, including stop word removal, stemming, and part-of-speech tagging. By leveraging the semantic understanding capabilities of large language models, we conduct deep semantic analysis on the descriptions of building material entries and extract key semantic features. Calculate the semantic matching score based on the key semantic features and the semantic features of the retrieval request; The comprehensive relevance score is dynamically adjusted based on market trends and user preferences for building materials.

[0015] Furthermore, it also includes: During the search process, if ambiguous or unclear statements are detected in the search request information, an interactive question-and-answer session with the user will be automatically triggered. Perform semantic analysis and intent recognition on user interactive questions and answers, and dynamically adjust the search keyword set and expanded keyword set based on user responses; The final search results are dynamically adjusted based on user feedback. Data mining analysis is performed on user search behavior and feedback to extract user search preferences and behavioral patterns.

[0016] The embodiments of the present invention have at least the following beneficial effects: 1. By utilizing a pre-trained large language model to perform deep semantic parsing on building material retrieval requests, key information such as the name, specifications, performance indicators, and application scenarios of building materials can be accurately extracted, generating a set of search keywords. This process not only understands the semantic information in the natural language descriptions input by users but also effectively handles fuzzy queries and multi-condition comprehensive searches. It solves the problem of inaccurate search results caused by the inability of traditional retrieval methods to accurately understand user intent, thus improving the accuracy and efficiency of retrieval.

[0017] 2. Based on the attribute characteristics and semantic relevance of building materials, the search keyword set is expanded in multiple dimensions, including synonym replacement, hyponym / hypernym expansion, industry terminology matching, and semantic association word generation, resulting in a more comprehensive and richer set of expanded keywords. In the building materials database, multi-dimensional searches are performed based on the expanded keyword set, comprehensively considering the name, specifications, performance indicators, manufacturers, and price attributes of building materials, yielding more comprehensive and accurate preliminary search results. This process solves the problem of traditional search methods performing poorly when handling multi-dimensional searches, meeting users' high requirements for the quality and comprehensiveness of search results.

[0018] 3. The preliminary search results are evaluated using multi-factor relevance scoring, including semantic matching score, attribute matching score, user evaluation score, and market trend score. The preliminary search results are then ranked based on the comprehensive relevance score, and building material entries with scores exceeding a preset threshold are selected as the final search results. Furthermore, the final search results are output to the user interface in the form of visual charts, with interactive functions and search result export functionality provided. This process not only comprehensively considers multiple factors to provide more accurate search results but also enhances the user experience through visual charts and interactive functions, addressing the shortcomings of traditional search methods in result presentation and user experience. Attached Figure Description

[0019] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein: Figure 1 This is a flowchart illustrating a building material retrieval method according to an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Figure 1 As shown, this application proposes a method for retrieving building materials, including: S1. Obtain building material retrieval request information, wherein the building material retrieval request information includes a natural language description input by the user; S2. Use a pre-trained large language model to perform deep semantic parsing on the building material retrieval request information, extract key information from the retrieval request, and generate a set of retrieval keywords. The key information includes the name, specifications, performance indicators, and application scenarios of the building materials. S3. Based on the building material attribute characteristics and semantic relevance, the search keyword set is expanded in multiple dimensions to generate an expanded keyword set. The multi-dimensional expansion includes synonym replacement, hyponym / hypernym expansion, industry terminology matching, and semantic association word generation. S4. In the building materials database, perform multi-dimensional retrieval based on the extended keyword set. The multi-dimensional retrieval includes a comprehensive search of the building materials' name, specifications, performance indicators, manufacturers, and price attributes to obtain preliminary search results. S5. Perform multi-factor relevance scoring on the preliminary search results. The multi-factor relevance scoring includes semantic matching score, attribute matching score, user evaluation score, and market trend score. Sort the preliminary search results according to the comprehensive relevance score to obtain the final search results. S6. Output the final search results to the user interface in the form of a visual chart, including bar charts, line charts and pie charts.

[0021] Building material retrieval requests refer to user descriptions of their building material needs input in natural language, specifically through text input or speech-to-text conversion. These requests capture the user's actual needs and transform them into processable search criteria. Pre-trained large language models are deep learning models trained on massive amounts of text data, such as BERT or GPT series models. These models perform deep semantic parsing of natural language descriptions and extract key information. The search keyword set is a collection of key information extracted from user input, typically achieved using named entity recognition and keyword extraction techniques. This set accurately reflects the core elements of the user's needs.

[0022] This application combines the deep semantic parsing capabilities of a large language model with multi-dimensional extended retrieval technology to construct a comprehensive retrieval system covering multiple attributes such as building material names, specifications, performance indicators, and application scenarios. It also introduces a multi-factor scoring mechanism to achieve accurate ranking of retrieval results, thereby solving the technical defects of traditional methods in handling natural language semantics and multi-dimensional related retrieval.

[0023] Multi-dimensional expansion refers to semantic expansion of the keyword set. This can be achieved through synonym replacement, hyponym / hypernym expansion, industry terminology matching, and semantic association word generation techniques, covering more potentially relevant search results and improving search comprehensiveness. A building materials database refers to a system storing structured data on building materials, which can be implemented using relational or NoSQL databases to support multi-dimensional indexing and fast retrieval. Multi-dimensional retrieval refers to comprehensive queries based on name, specifications, performance indicators, manufacturers, and price attributes. This can be achieved using inverted indexes and join query techniques to improve the coverage and accuracy of search results. Multi-factor relevance scoring refers to a comprehensive evaluation combining semantic matching degree, attribute matching degree, user reviews, and market trends. This can be achieved using weighted summation algorithms and logistic regression models to quantify the relevance of search results to user needs and achieve accurate ranking. Visual charts refer to displaying search results graphically, using data visualization tools such as bar charts, line charts, and pie charts to intuitively present the distribution characteristics of search results and comparisons of key indicators.

[0024] In implementation, the process begins by acquiring the building material retrieval request information, including the user's natural language description. Then, a pre-trained large language model is used to perform deep semantic parsing on the retrieval request information, extracting key information and generating a set of search keywords. This key information includes the building material's name, specifications, performance indicators, and application scenarios. The large language model, through deep learning technology, performs semantic understanding of natural language text, accurately capturing implicit information in the user's description.

[0025] Next, the search keyword set is expanded in multiple dimensions based on the attribute characteristics and semantic relevance of building materials, generating an expanded keyword set. This multi-dimensional expansion includes synonym replacement, hyponym / hypernym expansion, industry terminology matching, and semantic association word generation. This step significantly broadens the search scope by leveraging professional knowledge and semantic connections within the building materials field.

[0026] In the building materials database, multi-dimensional searches are performed based on an expanded set of keywords. These multi-dimensional searches include a comprehensive search of building materials' names, specifications, performance indicators, manufacturers, and price attributes, yielding preliminary search results. This multi-dimensional search method comprehensively considers various attributes of building materials, improving the accuracy and comprehensiveness of the search.

[0027] The preliminary search results are evaluated using a multi-factor relevance scoring system. This system includes semantic matching score, attribute matching score, user review score, and market trend score. The preliminary search results are then ranked based on the overall relevance score to obtain the final search results. This multi-factor scoring mechanism comprehensively considers various aspects of building materials, ensuring the relevance and usability of the search results.

[0028] Finally, the final search results are output to the user interface in the form of visual charts. These charts include bar charts, line charts, and pie charts, and this intuitive presentation helps users quickly understand and compare the search results.

[0029] In some embodiments, the user enters a search request through the system interface: a lightweight thermal insulation material suitable for the exterior walls of high-rise buildings is required, which has good fire resistance and a moderate price.

[0030] The system invokes a pre-trained large language model to perform deep semantic parsing on the input. The large language model identifies key information: the building material type is thermal insulation material, the application scenario is the exterior wall of a high-rise building, the performance requirements include lightweight, thermal insulation, and fire resistance, and the price requirement is moderate. The system generates an initial set of search keywords: [thermal insulation material, high-rise building, exterior wall, lightweight, fire resistance, moderate price].

[0031] The system expands the initial set of search keywords in multiple dimensions. Through synonym replacement, "lightweight" is expanded to "[lightweight, low density]"; through hyponym expansion, "insulation material" is expanded to "[insulation material, rock wool, glass wool, polystyrene]"; through industry terminology matching, "[Class A fireproof, thermal conductivity, density]" is added; and through semantic association word generation, "[energy saving, heat insulation]" is added.

[0032] The system performs multi-dimensional searches in the building materials database. Search criteria include: material type is thermal insulation material, application scenarios include high-rise buildings and exterior walls, and density is less than 0.8 g / cm³. 3 The thermal conductivity is less than 0.05 W / (m·K), the fire rating is Class A, and the price is in the medium range. The system obtained preliminary search results, including a variety of insulation materials that meet the criteria.

[0033] The system performs a multi-factor relevance score on the preliminary search results. The semantic matching score considers the semantic similarity between the material description and the search request; the attribute matching score considers the degree of matching between the material's performance indicators and the requirements; the user evaluation score considers historical user reviews of the material; and the market trend score considers the application trends of the material in the construction industry. The system then sorts the results based on the comprehensive score to obtain the final search results.

[0034] The system outputs the final search results to the user interface in the form of visual charts. A bar chart is used to display the overall score comparison of different materials, a line chart is used to display material price trends, and a pie chart is used to display the market share of different types of insulation materials.

[0035] This application further proposes to use a pre-trained large language model to perform deep semantic understanding on building material retrieval request information, extract key information as a set of retrieval keywords; normalize the set of retrieval keywords to remove duplicate and redundant keywords; and perform semantic enhancement on the set of retrieval keywords, generating semantic feature vectors related to the retrieval keywords through the large language model.

[0036] Deep semantic understanding identifies entities, attributes, and relationships in natural language descriptions through the attention mechanism of large language models. For example, models using the BERT architecture encode input text and output word vectors containing contextual information. Normalization processing, based on stemming and lemmatization techniques, converts words with different forms into a unified form; for example, merging concrete strength and high-strength concrete into concrete strength. Semantic enhancement maps keywords to a high-dimensional semantic space, generating vector representations with contextual relevance. For instance, the RoBERTa model generates 768-dimensional semantic feature vectors to capture potential associations between keywords.

[0037] Furthermore, the large language model encodes the natural language description input by the user, capturing deep semantic relationships in the text through a multi-layer Transformer structure, such as identifying the equivalence between compressive strength and load-bearing capacity. During the normalization stage, a hash table is used to store the processed keywords. When duplicate words or words with the same root are detected, a merging operation is automatically triggered, for example, unifying the C30 label and C30 grade to C30. In the semantic enhancement process, the large language model performs secondary encoding on the normalized keywords, generating semantic vectors with industry characteristics, such as generating composite vectors containing fire resistance time, temperature threshold, and application scenarios for fire-retardant coatings. These vectors are matched with database entries through cosine similarity calculation, improving the coverage and accuracy of semantic retrieval.

[0038] In some embodiments, a pre-trained large language model is used to perform deep semantic understanding of building material retrieval request information. The large language model adopts the BERT architecture, which encodes the input text through a multi-layer bidirectional Transformer encoder. The model input is a natural language description input by the user, which is fed into the BERT model after preprocessing steps such as word segmentation and word embedding. The BERT model outputs context-sensitive word vector representations of the text.

[0039] Based on the word vector representation output by BERT, a named entity recognition algorithm is used to extract key information from the search request, including the name, specifications, performance indicators, and application scenarios of building materials. The extracted key information constitutes a set of search keywords.

[0040] The set of search keywords is normalized. First, duplicate keywords are removed, and then synonyms are merged. For example, "rebar" and "rebar steel" are merged into "rebar". Redundant modifiers, such as "high-quality" and "excellent", are removed.

[0041] Semantic enhancement is performed on the search keyword set. A 300-dimensional word vector is generated for each keyword using the Word2Vec model. By calculating the cosine similarity of the word vectors, the five most relevant words for each keyword are identified and added to the search keyword set. Finally, the expanded search keyword set is deduplicated.

[0042] This application further proposes a multi-dimensional expansion of the search keyword set based on the attribute characteristics and semantic relevance of building materials, generating an expanded keyword set, including: performing synonym replacement and hyponym expansion on each search keyword in the search keyword set based on the attribute characteristics of building materials; generating relevant expanded keywords for each search keyword using semantic relevance analysis of a large language model; expanding the expanded keyword set by combining a professional terminology dictionary of the building materials industry; and deduplicating and filtering the expanded keyword set.

[0043] Synonym replacement is achieved through a predefined thesaurus of building material attributes or by generating a large language model, such as replacing compressive strength with compressive strength; hyponym expansion is based on the building material classification system, such as expanding cement to silicate cement or ordinary silicate cement; semantic relevance analysis uses a large language model to calculate the cosine similarity between keywords, and sets a similarity threshold to filter out expanded words with a relevance higher than 0.7; the professional terminology dictionary contains more than 5,000 standard terms in the building material industry, such as slump and water-cement ratio; deduplication filtering eliminates duplicates through a hash algorithm and retains expanded words with a frequency higher than a preset value based on word frequency statistics.

[0044] Furthermore, the search keywords are first analyzed for attribute features. For example, when the input keyword is "high-strength concrete," synonym replacement generates "high-strength concrete," and hypernym expansion generates "C60 concrete" and "prestressed concrete." Next, a large language model is used to analyze the semantic associations of "high strength," generating expanded terms such as compressive strength ≥50MPa and durability grade ≥P8. Further, a professional terminology dictionary is used to match industry standard parameters such as water-cement ratio ≤0.35 and chloride ion diffusion coefficient. Finally, a hash table is used to remove duplicate compressive strength entries, and expanded terms with a usage frequency higher than 10 times per month are selected based on historical search data.

[0045] In some embodiments, based on the characteristics of building material properties, each search keyword in the search keyword set is subjected to synonym replacement and hypernym expansion. For example, for the search keyword "concrete", synonym replacement yields "cement concrete" and "concrete", while hypernym expansion yields "ordinary concrete", "lightweight concrete", and "high-strength concrete", etc.

[0046] By leveraging the semantic relevance analysis of a large language model, relevant extended keywords are generated for each search keyword. Furthermore, the search keywords are input into a pre-trained large language model, which, through its semantic understanding capabilities, generates related extended keywords. For example, for the search keyword "insulation materials," extended keywords such as "heat insulation," "energy saving," and "wall insulation" are generated.

[0047] The set of extended keywords is expanded by combining them with a professional terminology dictionary for the building materials industry. Furthermore, the extended keywords are matched with the dictionary to add relevant professional terms. For example, for the extended keyword "waterproofing," professional terms such as "waterproof membrane," "waterproof coating," and "waterproof mortar" are added.

[0048] The expanded keyword set is deduplicated and filtered. This involves removing duplicate keywords and selecting the most relevant expanded keywords based on relevance scores, forming the final expanded keyword set.

[0049] This application further proposes a multi-dimensional search method in a building materials database based on an extended keyword set to obtain preliminary search results. This includes: indexing building materials entries in the database using multiple dimensions, including name index, specification index, performance index, manufacturer index, and price index; performing searches in each dimension of the index based on the extended keyword set to obtain multi-dimensional search results; merging and deduplicating the multi-dimensional search results to obtain preliminary search results; and performing preliminary screening on the preliminary search results to remove building materials entries that do not meet the search requirements.

[0050] The multi-dimensional index is constructed by creating independent index structures for building material entries based on name, specifications, performance indicators, manufacturers, and prices. Each index employs inverted index technology for fast keyword matching. For example, the name index uses a hash table to store the mapping between building material names and entry identifiers, while the performance indicator index is built as a range index segmented by numerical range. During the multi-dimensional retrieval process, keywords in the expanded keyword set are matched against the keyword fields of each index, generating multiple subsets from the matching results. Merging and deduplication operations are implemented using set operations, such as taking the union of multiple subsets and removing duplicate entries. Initial filtering is based on preset filtering conditions, such as excluding entries whose prices exceed the user's budget or whose performance indicators do not meet the standards.

[0051] Furthermore, the construction of multi-dimensional indexes involves splitting the entry attributes in the building materials database into independent dimensions, establishing a corresponding index structure for each dimension. For example, the name index uses an inverted index to store the mapping relationship between name keywords and entry identifiers, while the specification index uses a tree structure to store the range distribution of specification parameters. During the dimensional retrieval stage, keywords from the expanded keyword set are input into each dimension index. For instance, inputting "compressive strength ≥ 50MPa" into the performance index matches all entries with a compressive strength greater than or equal to 50MPa. The merging operation performs a union operation on the matching result sets of each dimension. For example, if the name dimension matches entries A and B, and the specification dimension matches entries B and C, the merged result is entries A, B, and C. The deduplication operation records the identifiers of merged entries using a hash table to avoid duplicate counting. Preliminary filtering involves traversing the merged entry set and filtering entries that do not meet the conditions according to preset rules, such as deleting entries whose manufacturers are not in the user-specified list. Through multi-dimensional indexing and step-by-step processing, retrieval efficiency is significantly improved and redundant results are reduced.

[0052] In some embodiments, the building material entries in the building materials database are indexed in multiple dimensions. These multiple dimensions include a name index, a specification index, a performance index, a manufacturer index, and a price index. The name index uses an inverted index structure, the specification and performance indexes use B+ tree structures, the manufacturer index uses a hash table structure, and the price index uses a skip list structure.

[0053] Based on the expanded keyword set, searches are performed across various index dimensions to obtain multi-dimensional search results. For the name index, a Boolean search model is used; for the specification and performance indexes, range queries are used; for the manufacturer index, exact matching is used; and for the price index, range queries are used.

[0054] The multi-dimensional search results are merged and deduplicated to obtain preliminary search results. The merging process uses a multi-way merge algorithm, and the deduplication process uses a bitmap filtering algorithm.

[0055] The initial search results undergo preliminary screening to remove building material entries that do not match the search query. Screening criteria include keyword matching, attribute consistency, and data completeness. Entries with a matching score below the threshold, inconsistent attributes, or incomplete data are discarded.

[0056] This application further proposes to perform multi-factor relevance scoring on each building material item in the preliminary search results, including semantic matching score, attribute matching score, user evaluation score and market trend score, and to obtain a comprehensive relevance score by weighted summation of each score, and to sort and select items with scores higher than a preset threshold as the final search results.

[0057] The semantic matching score extracts semantic feature vectors from the retrieval request and material description using a pre-trained large language model. Cosine similarity is calculated and adjusted using an sigmoid function, with the β coefficient controlling the steepness of the scoring curve. The attribute matching score is based on the degree of matching between the material item's attributes and the retrieval request, such as the conformity of specifications and performance indicators. The user evaluation score integrates historical user feedback data to reflect actual user satisfaction with the material's application. The market trend score combines real-time market data to analyze the material's supply and demand dynamics and industry trends. The weights of each score are dynamically configured according to the actual application scenario. For example, in scenarios emphasizing semantic matching, the semantic matching score weight is set to 0.5, the attribute matching score to 0.3, the user evaluation score to 0.1, and the market trend score to 0.1. A preset threshold of 0.7 is set to ensure high relevance of the final results.

[0058] Furthermore, when a user searches for lightweight partition wall panels with excellent fire resistance, the system first extracts the requested semantic vector vq. For a gypsum board item, its descriptive semantic vector vd is extracted, the cosine similarity is calculated to be 0.85, β is set to 1.2, resulting in a semantic score of 0.82. This material has a fire resistance rating of B1, matching the requested excellent requirement by 80%, with an attribute score of 0.8. The average user rating is 4.5 stars, converted to a score of 0.9. Current market trends show a 20% increase in demand for lightweight partition wall materials, resulting in a score of 0.85. The weighted sum is 0.82×0.5+0.8×0.3+0.9×0.1+0.85×0.1=0.828, exceeding the threshold of 0.7, and is included in the final results. Through this four-dimensional scoring mechanism, the system ensures both the matching of the material's essential attributes and considers user feedback and market dynamics, ensuring that the ranking results simultaneously meet technical indicators and commercial application needs.

[0059] In some embodiments, a semantic matching score is assigned to each building material item in the initial search results. The semantic matching score is obtained by calculating the semantic similarity between the search request and the description of the building material item. Further, a pre-trained BERT model is used to encode the search request and the description of the building material item respectively, obtaining corresponding semantic feature vectors. Further, the cosine similarity between the two semantic feature vectors is calculated to obtain a semantic similarity value. Finally, the semantic similarity value is substituted into the formula to calculate the semantic matching score.

[0060] Each building material entry in the initial search results is scored for attribute matching. The attribute matching score is calculated based on the degree of matching between the building material's specifications, performance indicators, and other attributes and the search request. For example, for numerical attributes, the relative error between the attribute value and the search request is calculated; for categorical attributes, it is determined whether there is a complete match or a partial match. The matching scores of each attribute are weighted to obtain the final attribute matching score.

[0061] Each building material item in the initial search results is assigned a user rating. These ratings are calculated based on historical user reviews of the material. Further, user star ratings and review texts are collected and weighted to obtain a comprehensive user rating score. Sentiment analysis is used to determine positive or negative emotions expressed in the review texts, which influence the final score as a weighting factor.

[0062] Each building material item in the initial search results is assigned a market trend score. This score is calculated based on market data such as historical sales volume and price trends. Furthermore, recent sales growth rates and price fluctuations are analyzed, and combined with industry development trends, to comprehensively assess the market prospects of the building materials and derive the final market trend score.

[0063] The semantic matching score, attribute matching score, user evaluation score, and market trend score are weighted and summed to obtain the comprehensive relevance score for each building material item. The weights can be adjusted according to specific application scenarios and user needs.

[0064] The initial search results are sorted in descending order based on a comprehensive relevance score. A scoring threshold is set, and building material entries with scores higher than the threshold are selected as the final search results. The threshold can be dynamically adjusted according to actual needs to balance the quantity and quality of search results.

[0065] This application further proposes to output the final search results to the user interface in the form of visual charts, including bar charts, line charts, and pie charts; to provide a detailed information page for the search results, including a detailed description of the building materials, manufacturer information, and user reviews; to provide interactive functions in the user interface, allowing users to further filter and sort the search results based on attributes; to generate dynamic interactive visual charts; and to provide an export function for the search results, supporting multiple formats.

[0066] The detailed information page displays detailed descriptions of building materials, manufacturer information, and user reviews in a structured manner, enabling users to quickly obtain key information. The interactive function allows users to further process search results based on specific attributes by setting filter conditions and sorting rules. Dynamic interactive visualization charts reflect the data distribution of different attribute dimensions through real-time updated bar charts, line charts, and pie charts. The export function supports data export in multiple formats, making it convenient for users to use search results for subsequent analysis or report generation.

[0067] Furthermore, in the user interface, search results are first displayed in the form of dynamic interactive charts. Bar charts compare the performance indicators of different building materials, line charts show price trends, and pie charts display the distribution of application scenarios. When a user clicks on a specific data point in the chart, a detailed information page pops up, displaying a detailed description of the corresponding building material, manufacturer information, and user reviews. Users can set filtering conditions through interactive functions, such as limiting the price range to a specific interval or sorting by user review ratings in descending order. The system dynamically updates the charts and lists based on the user's input. The dynamic interactive visualization charts are rendered in real-time using a real-time rendering engine, immediately redrawing the charts after the user adjusts the filtering conditions to maintain the real-time presentation of data. The export function converts the search results in the current view into table, image, or document formats by calling the format conversion module and generates download links for users to save. Through the combination of multi-dimensional visualization and interactive operations, users can quickly identify building materials that meet their needs and achieve efficient utilization of search results through data export.

[0068] In some embodiments, the final search results are output to the user interface in the form of visual charts. The detailed information page for the search results includes a detailed description of the building materials, manufacturer information, and user reviews. The user interface provides interactive features, allowing users to further filter and sort the search results based on attributes. Dynamic, interactive visual charts are generated. The search results can be exported in multiple formats.

[0069] Furthermore, the detailed information page for the search results uses a column layout. The left side displays basic information such as the name, specifications, and performance indicators of the building materials, while the right side displays manufacturer information and user reviews. The user review section includes rating statistics and text comments. Interactive functions are implemented through drop-down menus and sliders, allowing users to sort by price, performance indicators, or user ratings, and also set filtering conditions such as price range and performance indicator thresholds.

[0070] Dynamic, interactive visualizations include bar charts, line charts, and pie charts. Bar charts are used to compare the performance indicators or prices of different building materials, line charts display historical price trends for building materials, and pie charts show the proportion of different types of building materials in search results. These charts support interactive operations such as zooming, panning, and hovering over data points to display detailed information.

[0071] The search results export function supports multiple formats, including PDF, Excel, and CSV. PDF format retains the style of visual charts and is suitable for report generation. Excel and CSV formats facilitate further data analysis and processing. The export function allows users to choose to export all results or only the filtered results.

[0072] This application further proposes that the building materials database includes entries for building materials such as name, specifications, performance indicators, manufacturers, and prices; the building materials database is updated regularly; the building materials database supports dynamic expansion; the data in the building materials database undergoes quality control; the building materials database adopts a distributed storage architecture; the building materials database integrates a data mining module; and the building materials database supports multilingual retrieval.

[0073] Building material entries include name, specifications, performance indicators, manufacturer, and price, storing multi-dimensional attribute data through structured fields. A periodic update mechanism incrementally or fully updates the database content based on preset cycles or triggered events; the update cycle can be set weekly or monthly. Dynamic expansion allows for the addition of new building material entries via API or manual input, with new entries subject to format validation. The quality control process includes data integrity verification, logical consistency checks, and anomaly filtering; for example, performance indicator values ​​exceeding industry standard ranges are automatically marked as abnormal data. The distributed storage architecture uses sharding technology to distribute data across multiple physical nodes, each with an independent data read / write module. The data mining module incorporates clustering algorithms and association rule analysis models, periodically generating material attribute association maps. Multilingual search functionality is achieved through a built-in multilingual thesaurus, containing a bilingual (Chinese and English) glossary of building material terms.

[0074] Furthermore, building material entries are stored using structured fields, enabling the retrieval system to quickly match attribute fields such as name and specifications. A regular update mechanism ensures the database content is synchronized with market changes; for example, when new environmentally friendly materials are launched, the database automatically includes their parameters during the update cycle. Dynamic expansion allows the database to continuously grow with industry development; for example, when a new fireproof material category is added, relevant entries are imported in batches via an API interface. The quality control process eliminates errors through data cleaning rules, such as automatically correcting inconsistent unit values ​​in the specification field. The distributed storage architecture improves the speed of retrieving massive amounts of data through parallel computing; for example, when processing multiple retrieval requests simultaneously, different nodes execute retrieval tasks separately. The data mining module optimizes retrieval strategies by analyzing historical retrieval records; for example, when users frequently query seismic performance, the weight of this attribute in the scoring model is automatically increased. The multilingual retrieval function converts Chinese search terms into English terms through a terminology mapping table; for example, concrete is automatically mapped to "concrete" for cross-language retrieval.

[0075] In some embodiments, building material entries in the building materials database include the name, specifications, performance indicators, manufacturer, and price of the building materials. The building materials database is updated regularly. The building materials database supports dynamic expansion. The data in the building materials database undergoes quality control. The building materials database employs a distributed storage architecture. The building materials database integrates a data mining module. The building materials database supports multilingual retrieval.

[0076] Furthermore, the building materials database employs a distributed storage architecture, distributing data across multiple server nodes. Each node is responsible for storing a portion of the building materials data, enabling fast retrieval through a distributed index. The database undergoes periodic incremental updates, with newly added or modified building materials information being updated to all nodes in real time via a data synchronization mechanism.

[0077] The database supports dynamic expansion, allowing new data fields or attributes to be added as needed. For example, environmental certification information and lifespan of building materials can be added. Data quality control is conducted through a combination of automated testing and manual review to ensure data accuracy and consistency.

[0078] The data mining module integrates machine learning algorithms to analyze user search behavior and building material usage trends, providing support for ranking and recommending search results. The multilingual search function supports building material names and descriptions in multiple languages, including Chinese and English, enabling cross-language searches.

[0079] This application further proposes to utilize the context-aware capabilities of a large language model to generate context-related extended keywords for each search keyword in the search keyword set, to perform semantic verification on each extended keyword in the extended keyword set, and to dynamically optimize the extended keyword set based on user feedback and historical search data.

[0080] Context-aware capabilities analyze the complete semantic context of the search request using a large language model, identifying the implicit associations of search keywords in specific contexts and generating extended keywords consistent with the contextual logic. Semantic verification employs a pre-trained semantic similarity model to calculate the vector similarity between the extended keywords and the original search keywords, filtering out extended keywords with similarity scores below a preset threshold. Dynamic optimization establishes the association weights between extended keywords and users' actual needs by recording user click behavior and feedback ratings on search results, and periodically adjusts the extended keyword generation strategy.

[0081] Furthermore, when generating extended keywords, the large language model first parses the complete semantic structure of the search request, identifying the logical relationships and application scenarios between keywords. For example, when the search request includes "high-strength external wall insulation material," the model combines the context to identify the association between high strength and compressive strength and thermal conductivity, generating extended keywords such as compressive strength ≥50MPa and thermal conductivity ≤0.03W / (m·K). During the semantic verification stage, cosine similarity calculation is used. When the vector similarity between the extended keyword "fire resistance rating A" and the original keyword "insulation material" is less than 0.7, the extended keyword is automatically excluded. During dynamic optimization, the system statistically analyzes the click-through rate of search results containing "graphite polystyrene board." When the click-through rate of this keyword decreases by more than 15% for three consecutive months, its priority in the extended keyword set is reduced. Through these technical means, the contextual relevance and semantic accuracy of extended keywords are effectively improved, while simultaneously achieving adaptive optimization of the retrieval system.

[0082] In some embodiments, when a user inputs "fireproof and soundproof board" as a search keyword, the large language model analyzes the contextual semantic environment of the keyword to identify the fire resistance requirements and sound insulation performance indicators in the building curtain wall construction scenario, generating extended keywords such as "Class A fireproof rock wool board" and "composite sound-absorbing gypsum board." Furthermore, the generated lightweight fireproof partition wall system undergoes semantic verification, confirming that the term belongs to the standard terminology by searching the building materials industry standard database. Analysis of user click behavior data reveals that the click-through rate of "fireproof rock wool decorative integrated board" is consistently higher than other extended keywords. The system automatically increases the weight of this term in the extended keyword set, while removing "inorganic fireproof and soundproof coating," which has not been triggered for three months, from the extended keyword database.

[0083] This application further proposes to utilize the context-aware capabilities of a large language model to generate context-related extended keywords for each search keyword in the search keyword set, to perform semantic verification on each extended keyword in the extended keyword set, and to dynamically optimize the extended keyword set based on user feedback and historical search data.

[0084] Context-aware capabilities analyze the semantic structure of search requests using a large language model, identifying logical connections and application scenarios between keywords. For example, when a user inputs "high-strength exterior wall insulation material," the system generates extended terms such as "Class A fireproof rock wool board" and "polyurethane composite board," matching the building exterior wall scenario, based on the context. Semantic verification uses vector similarity calculation, comparing the semantic feature vectors of the extended keywords with the vectors of the original search request. Extended keywords are automatically filtered when the cosine similarity is below 0.7. Dynamic optimization establishes an extended keyword effectiveness evaluation model by recording user click behavior and feedback ratings on search results. For instance, when a user repeatedly ignores search results containing "extruded polystyrene board," the weight of that extended keyword in similar searches is reduced.

[0085] Furthermore, in the stage of generating expanded keywords, the large language model first parses the implicit semantics in the search request. For example, it maps the "humid environment" in waterproof materials suitable for humid environments to specific scenarios such as basements and bathrooms, thereby generating expanded terms with scenario adaptability, such as polymer self-adhesive film and cement-based penetrating crystallization. The semantic verification stage adopts a dual-threshold mechanism. When the semantic similarity between the expanded terms and the original keywords is in the range of 0.5-0.7, a manual review process is triggered to ensure the accuracy of professional terminology.

[0086] In some embodiments, during the preliminary result processing stage of the building material retrieval system, the descriptive text of each building material entry is preprocessed. The preprocessing includes word segmentation using natural language processing tools, filtering irrelevant words using a pre-defined stop word list, unifying word morphology using a lemmatization algorithm, and identifying noun phrases using part-of-speech tagging. The preprocessed text is input into a pre-trained large language model. The model analyzes the deep semantics of the text through an attention mechanism, extracting key semantic feature vectors containing material properties, application conditions, and performance parameters. Further, the model uses a bidirectional encoder structure to encode the text context, generating a 512-dimensional semantic representation vector. Further, the system calculates the cosine similarity between the semantic feature vector of the retrieval request and the feature vector of the building material entry to obtain a basic semantic matching score. Simultaneously, a market analysis module is integrated to obtain current quarterly building material price fluctuation data and a user preference model, incorporating market trend factors and user preference coefficients into the comprehensive score calculation through a linear weighting method.

[0087] This application further proposes that, during the retrieval process, if ambiguous or unclear expressions are detected in the retrieval request information, interactive question and answer with the user will be automatically triggered; semantic parsing and intent recognition will be performed on the user's interactive question and answer, and the retrieval keyword set and expanded keyword set will be dynamically adjusted according to the user's answers; the final retrieval results will be dynamically adjusted according to the user's feedback; and data mining analysis will be performed on the user's retrieval behavior and feedback to extract the user's retrieval preferences and behavioral patterns.

[0088] The system automatically triggers interactive question answering by identifying uncertain words or phrases in user input using a pre-set fuzzy detection algorithm, such as "similar," "approximately," or "applicable to a certain scenario." Semantic parsing and intent recognition employ a pre-trained large language model to perform contextual analysis on user responses and extract key limiting conditions. The system dynamically adjusts the search keyword set and expanded keyword set based on user-supplemented information, generating new keyword combinations through a semantic expansion algorithm. The final search results are also dynamically adjusted based on user feedback on initial results, such as marking irrelevant entries, and recalculating relevance scores. Data mining analysis uses clustering algorithms and association rule mining to identify high-frequency attributes and preference combinations in the user's historical searches.

[0089] Furthermore, when a user inputs a search for high-temperature resistant building materials, the system detects a lack of specific temperature ranges for "high-temperature resistant" and triggers an interactive query asking for specific temperature requirements. After the user answers that they need to withstand temperatures above 800 degrees Celsius, the large language model parses the key parameter 800 degrees Celsius and generates expanded keywords such as refractory materials and high-temperature resistant ceramics. The retrieval system adds these new keywords to the existing set, re-executes the search process, and filters out materials that meet the temperature requirements. The user marks alumina ceramics in the results as relevant, and the system records this preference, prioritizing the display of materials with similar properties in subsequent searches. By analyzing feedback data from multiple users, the system identifies that high-temperature environments are often associated with low thermal conductivity, and optimizes the expanded keyword generation strategy accordingly.

[0090] In some embodiments, during the operation of the building materials retrieval system, if a user inputs a search request containing fire-resistant materials suitable for high-rise buildings, the system detects ambiguity in the attribute indicators of the fire-resistant materials through a semantic confidence model. The interactive question-and-answer module is automatically activated, returning a dialog box containing radio buttons to the user. Options include professional indicators such as A1 fire resistance standard, B1 flame retardant rating, and fire resistance time ≥2 hours. After the user selects the A1 fire resistance standard, the system replaces the fire-resistant materials in the original search keyword set with A1 fire-resistant rock wool board and A1 fire-resistant gypsum board, and generates a combustion performance rating of A1 with a heat release rate ≤20kW / m² through semantic association. 2 Expanding keywords. During database retrieval, multi-dimensional matching is re-executed based on the updated keyword set, and user selection behavior is recorded in the behavior analysis log. When a user clicks to view the returned A1-grade fireproof rock wool board item, the system marks this behavior as positive feedback, dynamically increasing the weight coefficient of rock wool materials in subsequent search rankings.

[0091] Through the above technical solution, this application effectively solves the problem of retrieval result bias caused by fuzzy user queries. By using a context-aware interaction mechanism, unstructured natural language is transformed into structured search conditions, creating a closed-loop optimization process for generating the search keyword set. Simultaneously, an implicit feedback model is established based on user behavior data, enabling the system to adaptively adjust its search strategy and proactively avoid similar fuzzy expressions in subsequent interactions, thereby improving the matching accuracy between search results and users' actual needs.

[0092] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for retrieving building materials, characterized in that, Includes the following steps: Obtain building material retrieval request information, which includes a natural language description input by the user; The pre-trained large language model is used to perform deep semantic parsing on the building material retrieval request information, extract key information from the retrieval request, and generate a set of retrieval keywords. The key information includes the name, specifications, performance indicators, and application scenarios of the building materials. Based on the property characteristics and semantic relevance of building materials, the set of search keywords is expanded in multiple dimensions to generate an expanded set of keywords. The multi-dimensional expansion includes synonym replacement, hyponym and hypernym expansion, industry terminology matching, and semantic association word generation. In the building materials database, a multi-dimensional search is performed based on the extended keyword set. The multi-dimensional search includes a comprehensive search of the building materials' name, specifications, performance indicators, manufacturers, and price attributes to obtain preliminary search results. The preliminary search results are subjected to a multi-factor relevance score, which includes semantic matching score, attribute matching score, user evaluation score and market trend score. The preliminary search results are sorted according to the comprehensive relevance score to obtain the final search results. The final search results are output to the user interface in the form of visual charts, including bar charts, line charts, and pie charts.

2. The building material retrieval method according to claim 1, characterized in that, The method utilizes a pre-trained large language model to perform deep semantic parsing on the building material retrieval request information, extracts key information from the retrieval request, and generates a set of retrieval keywords, including: A pre-trained large language model is used to perform deep semantic understanding on the building material retrieval request information, and key information is extracted as a set of retrieval keywords. The set of search keywords is normalized to remove duplicate and redundant keywords; The set of search keywords is semantically enhanced by generating semantic feature vectors related to the search keywords through a large language model.

3. The building material retrieval method according to claim 1, characterized in that, The method of expanding the search keyword set in multiple dimensions based on the building material attribute characteristics and semantic relevance to generate an expanded keyword set includes: Based on the characteristics of building material properties, synonym replacement and hyponym expansion are performed on each search keyword in the search keyword set; Using semantic relevance analysis of a large language model, related extended keywords are generated for each search keyword; The extended keyword set was expanded by combining it with a professional terminology dictionary for the building materials industry; The extended keyword set is deduplicated and filtered.

4. The building material retrieval method according to claim 1, characterized in that, The preliminary search results, obtained by performing a multi-dimensional search based on the extended keyword set in the building materials database, include: The building material entries in the building materials database are indexed in multiple dimensions, including name index, specification index, performance index, manufacturer index, and price index. Based on the extended keyword set, searches are performed in the indexes of each dimension to obtain multi-dimensional search results; The search results from multiple dimensions are merged and deduplicated to obtain preliminary search results; The initial search results are initially filtered to remove building material entries that do not meet the search requirements.

5. The building material retrieval method according to claim 1, characterized in that, The process of performing multi-factor relevance scoring on the preliminary search results to obtain the final search results includes: For each building material item in the preliminary search results, a semantic matching score is calculated. Based on a large language model, the semantic similarity between the search request and the description of the building material item is calculated. The formula for calculating the semantic matching score is as follows: in, Indicates the semantic matching score. This represents the semantic feature vector of the retrieval request. The semantic feature vector representing the building material item. For adjustment coefficients; Each building material item in the preliminary search results is scored for attribute matching degree. Each building material item in the preliminary search results will be evaluated and rated by users. Each building material item in the preliminary search results will be scored for market trends. The semantic matching score, attribute matching score, user evaluation score, and market trend score are weighted and summed to obtain the comprehensive relevance score for each building material item. The preliminary search results are sorted based on the comprehensive relevance score, and building material entries with scores higher than a preset threshold are selected as the final search results.

6. The building material retrieval method according to claim 1, characterized in that, The step of outputting the final search results to the user interface in the form of a visual chart includes: Provide a detailed information page for the search results, which includes a detailed description of the building materials, manufacturer information, and user reviews; The user interface provides interactive features that allow users to further filter and sort search results based on attributes; Generate dynamic, interactive visualizations; The system provides a function to export search results, which supports multiple formats.

7. The building material retrieval method according to claim 1, characterized in that, The building material entries in the building materials database include the name, specifications, performance indicators, manufacturer, and price of the building materials; The building materials database is updated regularly. The building materials database supports dynamic expansion; The data in the building materials database undergoes quality control. The building materials database adopts a distributed storage architecture; The building materials database integrates a data mining module; The building materials database supports multilingual retrieval.

8. The building material retrieval method according to claim 1, characterized in that, The step of expanding the search keyword set in multiple dimensions based on building material attribute characteristics and semantic relevance to generate an expanded keyword set also includes: By leveraging the context-aware capabilities of large language models, context-related extended keywords are generated for each search keyword in the search keyword set; Perform semantic validation on each extended keyword in the extended keyword set; The expanded keyword set is dynamically optimized based on user feedback and historical search data.

9. The building material retrieval method according to claim 1, characterized in that, The step of performing multi-factor relevance scoring on the preliminary search results to obtain the final search results also includes: Each building material entry in the preliminary search results is preprocessed with text, including stop word removal, stemming, and part-of-speech tagging. By leveraging the semantic understanding capabilities of large language models, we conduct deep semantic analysis on the descriptions of building material entries and extract key semantic features. Calculate the semantic matching score based on the key semantic features and the semantic features of the retrieval request; The comprehensive relevance score is dynamically adjusted based on market trends and user preferences for building materials.

10. The building material retrieval method according to claim 1, characterized in that, Also includes: During the search process, if ambiguous or unclear statements are detected in the search request information, an interactive question-and-answer session with the user will be automatically triggered. Perform semantic analysis and intent recognition on user interactive questions and answers, and dynamically adjust the search keyword set and expanded keyword set based on user responses; The final search results are dynamically adjusted based on user feedback. Data mining analysis is performed on user search behavior and feedback to extract user search preferences and behavioral patterns.

Images