Intelligent retrieval method for engineering material price based on fine-grained field perception fusion

Abstract

The application discloses an intelligent retrieval method for engineering material price based on fine-grained field perception fusion, comprising the following steps: receiving an original query set and providing a material library; performing semantic alignment and structure perception enhancement processing on original queries in the original query set to generate normalized enhanced queries; calculating global similarity between the normalized enhanced queries and records in the material library through dense semantic channels and sparse term channels; performing field-level analysis on the normalized enhanced queries and each record in the material library to calculate field-level fine-grained similarity of the dense semantic channels and the sparse term channels on a pre-defined key field set; performing normalization processing on the global similarity and the field-level fine-grained similarity, performing weighted fusion, and generating final fusion similarity; and sorting the records in the material library according to the final fusion similarity, and outputting the record with the highest ranking as target material price information. The application effectively alleviates the influence of noise and fuzzy queries on matching accuracy.

Description

Technical Field

[0001] This invention relates to the field of engineering construction technology, and in particular to an intelligent retrieval method for engineering material prices based on fine-grained field perception fusion. Background Technology

[0002] With the increasing complexity of industrial scenarios and the rapid growth of data scale, the role of information retrieval and knowledge alignment in industrial applications is becoming increasingly prominent. Unlike traditional structured databases, equipment data in the engineering field often exists in the form of unstructured text, making traditional methods relying on Structured Query Language (SQL) ineffective in handling complex query requirements. This problem is particularly prominent in engineering material price retrieval: query text often contains noise, abbreviations, and vague expressions, directly leading to a decrease in retrieval accuracy and severely restricting the efficiency and reliability of project cost estimation. In project cost estimation applications in the construction industry, equipment specification information tables often contain complex semantic information. How to extract key features from them and achieve accurate matching is the core challenge of the retrieval task. Existing research has attempted to solve this problem from the perspective of multi-granularity semantic retrieval. For example, information intelligent retrieval methods based on multi-granularity semantic fusion improve matching accuracy by fusing semantic information of different granularities such as words, phrases, and text, or partial multimodal hashing methods based on fine-granular feature fusion, which use fine-granular feature fusion and self-attention mechanisms to mitigate the impact of missing multimodal features. However, these methods are still insufficient when faced with real queries containing noise, abbreviations, and vague expressions. On the other hand, research on price knowledge alignment shows that the introduction of quota intelligent binding and knowledge graphs can improve the matching efficiency of equipment specification information and quota price, but it is difficult to guarantee the stability and robustness of the matching results in complex semantic scenarios.

[0003] In recent years, information retrieval (IR) has made significant progress in areas such as search engines, question-answering systems, and personalized recommendations. Some studies have attempted to improve this by optimizing the index structure. For example, Cheng Yining et al. proposed "A Study on the Interactive Application of Index Structure Optimization-Based Retrieval Enhancement Generation Technology in the Insurance Field," which introduces document segmentation, keyword extraction, and semantic alignment mechanisms during the indexing process, thereby improving the ability to capture key information. This method has advantages in improving retrieval coverage, but its ability to handle semantic ambiguity and cross-field expressions remains insufficient.

[0004] In the field of semantic representation learning, Wang Yulong et al. proposed a method for extracting knowledge entities from academic documents by integrating semantic enhancement and knowledge distillation. This method combines global semantics with dynamic word meaning information by integrating the embedding representation methods of SciBERT and ELMo models, effectively improving the modeling ability of complex contexts. This method performs well in capturing deep semantics, but its ability to distinguish explicit elements such as numerical fields and unit specifications is limited, making it difficult to directly apply to engineering materials scenarios.

[0005] To integrate the advantages of sparse and dense models, Yang Dongju and Yang Kun proposed a "Domain Question Answering Alignment Optimization Method Based on Keyword Extraction and Hybrid Retrieval." This method utilizes a large language model to extract user query keywords and combines sparse and dense retrieval to achieve hybrid recall, thereby enhancing coverage and robustness. However, this method still relies on empirical settings for weight allocation and dynamic adaptability, making it difficult to achieve stable optimal fusion in complex scenarios.

[0006] In summary, although existing research has made positive progress in index optimization, sparse and dense fusion, semantic representation, and multi-task learning, there are still significant shortcomings in the scenario of engineering material price retrieval. Faced with the specialized engineering field containing noise, abbreviations, and ambiguous expressions, retrieval is prone to semantic drift; single modeling methods struggle to simultaneously address deep semantic understanding and explicit field matching; and insufficient fusion of features from different channels makes it difficult to balance recall coverage and discrimination accuracy in candidate ranking. Summary of the Invention

[0007] In view of this, the purpose of this invention is to propose an intelligent retrieval method for engineering material prices based on fine-grained field-aware fusion, which effectively mitigates the impact of noise and fuzzy queries on matching accuracy; designs a candidate generation mechanism with dense-sparse dual-channel dynamic fusion to improve recall coverage while ensuring semantic discrimination capability; and constructs a low-cost, scalable retrieval system with the potential for rapid application in the engineering field.

[0008] To achieve the above-mentioned technical objectives, the technical solution adopted by this invention is as follows: This invention provides an intelligent retrieval method for engineering material prices based on fine-grained field-aware fusion, comprising the following steps: Step 1: Receive the unstructured raw query set and provide a semi-structured material library; Step 2: Perform semantic alignment and structure-aware enhancement processing on the original queries in the original query set to generate normalized enhanced queries; Step 3: Calculate the global similarity between the normalized enhanced query and each record in the material database through both dense semantic channels and sparse term channels; Step 4: Perform field-level parsing on the normalized enhanced query and each record in the material library, and calculate the field-level fine-grained similarity of the dense semantic channel and the sparse term channel on the predefined key field set respectively; Step 5: Normalize the global similarity and field-level fine-grained similarity, obtain the global fusion weight and field-level fusion weight, perform weighted fusion, and generate the final fusion similarity. Step 6: Sort all records in the material library according to the final fusion similarity, and output the record with the highest ranking as the target material price information.

[0009] Furthermore, step 1 specifically includes: Step 11: Receive the unstructured raw query set for the engineering material price retrieval scenario. ,in, Indicates the first A raw query described in natural language. Indicates the index of the original query. This represents the total number of original queries, which include noise, industry abbreviations, and colloquial expressions. Step 12: Provide a semi-structured material library ,in, For structural patterns, For recording sets, This represents the j-th record, where j represents the index of the record, and n represents the total number of records. Each record... It consists of multiple heterogeneous fields, among which, This indicates the first heterogeneous field in the j-th record. This indicates the second heterogeneous field in the j-th record. This indicates the k-th heterogeneous field in the j-th record, where k represents the total number of heterogeneous fields; the heterogeneous fields include structured numeric / enumerated fields and free text fields.

[0010] Furthermore, step 2 specifically includes: Step 21: Perform semantic alignment processing on each original query in the original query set to generate semantically equivalent standardized queries; Step 22: Perform structure-aware enhancement processing on the standardized query to form a structured query field representation; Step 23: Concatenate the standardized query with the query field representation to generate a normalized enhanced query.

[0011] Furthermore, step 21 specifically includes: The semantic alignment process includes at least one of the following operations: synonym normalization, abbreviation expansion, and unit normalization. Through synonym normalization mapping, abbreviation expansion rule set, and unit normalization conversion function, noise, abbreviations, and fuzzy expressions in the original query are converted into standard forms, generating semantically equivalent standardized queries. ; Step 22 specifically includes: Step 221: Define the domain field set :

[0012] Among them, the domain field set is divided into materials according to the engineering domain. ,Specification and unit ; Step 222: Using rule matching, dictionary mapping, or numerical recognition methods, extract data from standardized queries. Extract predefined key fields and their corresponding values ​​to obtain a structured query field representation. :

[0013] in, Indicates key fields, Key fields The range of values, Key fields The value of ; Step 23 specifically includes: Standardized queries With structured query field representation Concatenating and serializing results in normalized enhanced queries. :

[0014] in, This is a concatenation function.

[0015] Furthermore, the aforementioned synonym normalization, abbreviation expansion, and unit normalization specifically include: Synonym normalization specifically involves: normalizing domain-specific vocabulary. abbreviations or aliases in Mapping to standard words Normalization yields vocabulary ; Abbreviation expansion utilizes rule sets. Replace the abbreviation 'a' with the standard form 'b'; Unit normalization is for numerical-unit pairs Using conversion functions This will be transformed into a unified measurement system.

[0016] Furthermore, step 3 specifically includes: Step 31: Construct a dense-sparse dual-channel framework, including a dense semantic channel and a sparse term channel; Step 32: In dense semantic channels, a pre-trained semantic encoder is used. Normalization Enhanced Queries and each record in the material library Mapping to query semantic vector and record vector : , and

[0017] in, express Normalization operation; , Generally refers to a pre-trained semantic encoder The encoded result; Represents the real number field; Represents a set of d-dimensional real vectors; Query enhancement through normalization using the cosine inner product of vectors. and each record in the material library The similarity is used to obtain the global dense semantic similarity. :

[0018] in, Represents the query semantic vector and record vector inner product operation; Step 33: In the sparse term channel, based on the word segmentation function The term set is obtained, and the normalized enhanced query is calculated using the BM25 model. With each record in the material library The term similarity between terms is used to obtain the global sparse term similarity. :

[0019] in, Representing words, For words In Record word frequency in This represents a hyperparameter that adjusts word frequency saturation; its function is to control word frequency. The degree of influence on the final relevance score; Used to control the strength of normalization; Representing records Length, This represents the average record length. The inverse recording frequency is defined as:

[0020] in, To record the total, For words containing The number of records.

[0021] Furthermore, step 4 specifically includes: Step 41: Parse the corresponding content of the same key field from each record in the normalized enhanced query and the material library respectively; Step 42: For each key field f, use the semantic encoder of the dense semantic channel to calculate the similarity between its field content vectors, and obtain the field-level dense similarity. ; Step 43: For each key field f, use the BM25 model of the sparse term channel to calculate the similarity between its field content terms, and obtain the field-level sparse term similarity. .

[0022] Furthermore, step 5 includes: Step 51: Apply the Min-Max normalization operator to the global dense semantic similarity. Global sparse term similarity Field-level dense similarity Similarity to field-level sparse terms Normalize it to unify its scale to the [0,1] interval:

[0023] in, express Normalization results express Normalization results express Normalization results express Normalization results; Step 52: Input the semantic vector of the normalized enhanced query into the trained multilayer perceptron model to predict the global fusion weight and multiple field-level fusion weights for the current query; Step 53: Use the global fusion weight to linearly weight the normalized global dense semantic similarity and the normalized global sparse term similarity to obtain the dual-channel fusion similarity of global semantics. :

[0024] in, For global fusion weights; Step 54: Apply the field-level fusion weights to linearly weight the normalized field-level dense similarity and normalized field-level sparse similarity of the corresponding fields to obtain the dual-channel fusion similarity of field-level semantics. :

[0025] in, Assign weights to each field level; Step 55: Fuse the dual-channel similarity of global semantics. Dual-channel fusion similarity with semantics at each field level Summing and dividing by the number of fields plus one yields the final fused similarity. :

[0026] in, The similarity is a dual-channel fusion of global semantics; This represents the dual-channel fusion similarity for field-level semantics.

[0027] Furthermore, the training process of the multilayer perceptron model includes: 1) Constructing training samples: For the i-th normalized enhanced query In the discrete weight set Enumeration weight value Choose the option that allows the true target to be recorded. Enter based on Optimal weights of the retrieved Top-K ranking candidate set As a reference value;

[0028] in, For indicator functions, Indicates the record being retrieved; Representation and Normalization Enhanced Queries The most similar One record; Similarly, for the i-th normalized enhanced query In the discrete weight set Enumeration weight value Choose the option that allows the true target to be recorded. Enter based on Optimal weights of the retrieved Top-K ranking candidate set As a reference value;

[0029] Obtain the adaptive weight training sample set for:

[0030] in, To query the training set; 1) Using the query semantic vector For input, a multilayer perceptron is used. Output fusion weights As a predicted value:

[0031] in, The neural network representing the multilayer perceptron maps d-dimensional vectors to prediction weights; Represents a set of d-dimensional real vectors; Similarly, with the query semantic vector For input, a multilayer perceptron is used. Output field-level fusion weights As a predicted value:

[0032] 3) During the training phase, the optimization objective is to minimize the mean squared error between the predicted and reference values, while introducing... The regularization term is used to train the multilayer perceptron model using the AdamW optimizer:

[0033] in, These represent the learning parameters of the weighted scoring model. This represents the loss function that guides the training of a neural network. The regularization coefficient is . express Regular terms.

[0034] Furthermore, during the inference phase, the fusion weights predicted by the multilayer perceptron model are... Perform range clipping operation:

[0035] in, This means truncating the weights to a closed interval. Inside, The value is selected as [0.7, 0.9].

[0036] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art: This invention proposes an intelligent retrieval method for engineering material prices based on fine-grained field-aware fusion, effectively alleviating the shortcomings in retrieval accuracy and stability caused by low retrieval efficiency, long processing time, and subjective human intervention. Through semantic alignment and structure-aware enhancement, it achieves standardized expression of non-standardized queries; by combining dense semantic channels and sparse term channels, it effectively balances semantic understanding and precise term matching. Simultaneously, through a dynamic weight learning mechanism, it achieves stable identification of target records under unique solution constraints. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is an execution flowchart of an intelligent retrieval method for engineering material prices based on fine-grained field perception fusion provided in an embodiment of the present invention.

[0039] Figure 2 This is a graph showing the impact of different weight parameters on the model retrieval performance, provided in an embodiment of the present invention. Detailed Implementation

[0040] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] Please see Figure 1 The present invention provides an intelligent retrieval method for engineering material prices based on fine-grained field-aware fusion, comprising the following steps: Step 1: Receive the unstructured raw query set and provide a semi-structured material library; In this embodiment, step 1 specifically includes: Step 11: Receive the unstructured raw query set for the engineering material price retrieval scenario. ,in, Indicates the first A raw query described in natural language. Indicates the index of the original query. This represents the total number of original queries, which include noise, industry abbreviations, and colloquial expressions. Step 12: Provide a semi-structured material library ,in, For structural patterns, For recording sets, This represents the j-th record, where j represents the index of the record, and n represents the total number of records. Each record... It consists of multiple heterogeneous fields, among which, This indicates the first heterogeneous field in the j-th record. This indicates the second heterogeneous field in the j-th record. This indicates that the j-th record contains the k-th heterogeneous field, where k represents the total number of heterogeneous fields. The heterogeneous fields include structured numerical / enumerated fields (such as units, specifications, years, etc.) and free text fields (such as project name, project characteristics, work content description, etc.).

[0042] Step 2: Perform semantic alignment and structure-aware enhancement processing on the original queries in the original query set to generate normalized enhanced queries; In the scenario of searching for engineering material prices, the original query These often contain noise, abbreviations, and domain-specific ambiguities. Directly using them for comparison can easily lead to semantic drift and make field alignment difficult. Therefore, this invention, while maintaining semantic equivalence, introduces query semantic enhancement and structure-aware modeling to standardize the input query, making core elements such as name, specifications, parameters, and units explicitly presented, thereby reducing semantic ambiguity and enhancing its comparability with structured fields. Specifically, semantic alignment and structure-aware enhancement operators are introduced: , in For the original query set, This is the normalized enhanced query set.

[0043] The original query is enhanced by semantic enhancement operators. Convert to normalized expression This generates representations that combine semantic fidelity and structural explicitness, laying a solid foundation for subsequent similarity modeling.

[0044] In this embodiment, step 2 specifically includes: Step 21: Perform semantic alignment processing on each original query in the original query set to generate semantically equivalent standardized queries; the goal of semantic alignment is to unify noise, abbreviations, and diverse colloquial expressions into a standard form, and this process yields standardized queries. Its semantic expression is equivalent to the original query, but its form is more standardized.

[0045] Preferably, step 21 specifically includes: The semantic alignment process includes at least one of the following operations: synonym normalization, abbreviation expansion, and unit normalization. Through synonym normalization mapping, abbreviation expansion rule set, and unit normalization conversion function, noise, abbreviations, and fuzzy expressions in the original query are converted into standard forms, generating semantically equivalent standardized queries. ; Specifically, the synonym normalization, abbreviation expansion, and unit normalization include: Synonym normalization specifically involves defining the synonym normalization mapping as follows: domain vocabulary abbreviations or aliases in Mapping to standard words ( ), normalization yields vocabulary ; Abbreviation expansion utilizes rule sets. Replace the abbreviation 'a' with the standard form 'b'; Unit normalization is for numerical-unit pairs (in, (where u is the unit) Use conversion functions This will be converted into a unified measurement system (e.g., "2cm" → "20mm").

[0046] Step 22: Perform structure-aware enhancement processing on the standardized query to form a structured query field representation; after obtaining the standardized query... Next, key fields need to be further specified to align with the database's structured schema.

[0047] Preferably, step 22 specifically includes: Step 221: Define the domain field set :

[0048] Among them, the domain field set is divided into materials according to the engineering domain. ,Specification and unit ; Step 222: Using rule matching, dictionary mapping, or numerical recognition methods, extract data from standardized queries. Extract predefined key fields and their corresponding values ​​to obtain a structured query field representation. :

[0049] in, Indicates key fields, Key fields The range of values, Key fields The value of ; Step 23: Concatenate the standardized query with the query field representation to generate a normalized enhanced query.

[0050] Preferably, step 23 specifically includes: Standardized queries With structured query field representation Concatenating and serializing results in normalized enhanced queries. :

[0051] in, This is a concatenation function.

[0052] Step 3: Calculate the global similarity between the normalized enhanced query and each record in the material database through both dense semantic channels and sparse term channels; In the candidate generation stage, single-channel similarity modeling often has limitations. While dense vectors generated through dense semantic channels can capture deep semantic relationships, they are insufficient when dealing with key fields of engineering materials such as specifications, values, and units. Sparse term channels, on the other hand, use term-based models (such as BM25) to provide accurate matching, but struggle to cover semantic synonyms and contextual dependencies. Therefore, to better adapt to the coexistence of semantic and numerical elements in engineering material retrieval, a field-aware fine-grained fusion mechanism is introduced on top of the traditional dense-sparse dual-channel framework.

[0053] In this embodiment, step 3 specifically includes: Step 31: Construct a dense-sparse dual-channel framework, including a dense semantic channel and a sparse term channel; Step 32: In dense semantic channels, a pre-trained semantic encoder is used. Normalization Enhanced Queries and each record in the material library Mapping to query semantic vector and record vector : , and

[0054] in, express Normalization operation; , Generally refers to a pre-trained semantic encoder The encoded result; Represents the real number field; Represents a set of d-dimensional real vectors; Query enhancement through normalization using the cosine inner product of vectors. and each record in the material library The similarity is used to obtain the global dense semantic similarity. :

[0055] in, Represents the query semantic vector and record vector inner product operation; Step 33: In the sparse term channel, based on the word segmentation function The term set is obtained, and the normalized enhanced query is calculated using the BM25 model. With each record in the material library The term similarity between terms is used to obtain the global sparse term similarity. :

[0056] in, Representing words, For words In Record word frequency in This represents a hyperparameter that adjusts word frequency saturation; its function is to control word frequency. The degree of influence on the final relevance score; Used to control the strength of normalization; Representing records Length, This represents the average record length. The inverse recording frequency is defined as:

[0057] in, To record the total, For words containing The number of records.

[0058] Step 4: Perform field-level parsing on each record in the normalized enhanced query and material library, and calculate the field-level parsing of the dense semantic channel and sparse term channel on the predefined key field set respectively. Fine-grained similarity; To further enhance adaptability to engineering material retrieval scenarios, a field-aware fine-grained fusion mechanism is introduced based on global dual-channel modeling. Traditional global weighted fusion is prone to the problem of "semantic matching is correct but field mismatch" when dealing with engineering records containing multiple heterogeneous elements: for example, querying "Φ20 steel bar" may still have a high overall similarity to "Φ25 steel bar, 20 tons", but ignores the inconsistency of the specification field. Therefore, this invention proposes dense-sparse dual-channel modeling and fusion at the field level to ensure accurate alignment of each key element.

[0059] In this embodiment, step 4 specifically includes: Step 41: Parse the corresponding content of the same key field from each record in the normalized enhanced query and the material library respectively; Step 42: For each key field f, use the semantic encoder of the dense semantic channel to calculate the similarity between its field content vectors, and obtain the field-level dense similarity. ; Step 43: For each key field f, use the BM25 model of the sparse term channel to calculate the similarity between its field content terms, and obtain the field-level sparse term similarity. .

[0060] Step 5: Normalize the global similarity and field-level fine-grained similarity, obtain the global fusion weight and field-level fusion weight, perform weighted fusion, and generate the final fusion similarity. In this embodiment, step 5 includes: Step 51: To eliminate the difference in numerical scale between dense semantic similarity and sparse term similarity, the global dense semantic similarity is normalized using the Min-Max normalization operator. Global sparse term similarity Field-level dense similarity Similarity to field-level sparse terms Normalize it to unify its scale to the [0,1] interval:

[0061] in, express Normalization results express Normalization results express Normalization results express Normalization results; Step 52: Input the semantic vector of the normalized enhanced query into the trained multilayer perceptron model to predict the global fusion weight and multiple field-level fusion weights for the current query; To balance the relative importance of semantic matching and term matching across different queries, a query-based dual-channel adaptive weighting mechanism is introduced. For each query... Learning a fusion weight Then, the normalized scores of the two channels are linearly fused.

[0062] Step 53: Use the global fusion weight to linearly weight the normalized global dense semantic similarity and the normalized global sparse term similarity to obtain the dual-channel fusion similarity of global semantics. :

[0063] in, For global fusion weights; Step 54: Apply the field-level fusion weights to linearly weight the normalized field-level dense similarity and normalized field-level sparse similarity of the corresponding fields to obtain the dual-channel fusion similarity of field-level semantics. :

[0064] in, Assign weights to each field level; Step 55: Fuse the dual-channel similarity of global semantics. Dual-channel fusion similarity with semantics at each field level Summing and dividing by the number of fields plus one yields the final fused similarity. :

[0065] in, The similarity is a dual-channel fusion of global semantics; This represents the dual-channel fusion similarity for field-level semantics.

[0066] The training process of the multilayer perceptron model includes: 1) Constructing training samples: For the i-th normalized enhanced query In the discrete weight set Enumeration weight value Choose the option that allows the true target to be recorded. Enter based on Optimal weights of the retrieved Top-K ranking candidate set As a reference value;

[0067] in, For indicator functions, Indicates the record being retrieved; Representation and Normalization Enhanced Queries The most similar One record; Similarly, for the i-th normalized enhanced query In the discrete weight set Enumeration weight value Choose the option that allows the true target to be recorded. Enter based on Optimal weights of the retrieved Top-K ranking candidate set As a reference value;

[0068] If all Obtain the adaptive weight training sample set for:

[0069] in, To query the training set; 2) Using the query semantic vector For input, a multilayer perceptron is used. Output fusion weights As a predicted value:

[0070] in, The neural network representing the multilayer perceptron maps d-dimensional vectors to prediction weights; Represents a set of d-dimensional real vectors; Similarly, with the query semantic vector For input, a multilayer perceptron is used. Output field-level fusion weights As a predicted value:

[0071] In actual reasoning, if the continuous weights output by the model are used directly... This can lead to extreme value biases, resulting in over-reliance on a single channel. This can amplify the error of a particular channel, causing instability in the sorting results. To suppress fluctuations caused by extreme values, during the inference phase, the fusion weights predicted by the multilayer perceptron model are adjusted. Perform range clipping operation:

[0072] in, This means truncating the weights to a closed interval. Inside, The value is selected as [0.7, 0.9].

[0073] 3) To make the predicted value It can approximate the optimal reference value obtained based on discrete search. Predicted value It can approximate the reference value This invention models the weight prediction problem as a regression task. During the training phase, the optimization objective is to minimize the mean squared error between the predicted and reference values, while simultaneously introducing... The regularization term is used to train the multilayer perceptron model using the AdamW optimizer:

[0074] in, These represent the learning parameters of the weighted scoring model. This represents the loss function that guides the training of a neural network. The regularization coefficient is . express Regular terms.

[0075] Step 6: Sort all records in the material library according to the final fusion similarity, and output the record with the highest ranking as the target material price information.

[0076] The objective of this invention is to target a specific original query. In the record set Select the unique target record :

[0077] in, This means for all; This represents the i-th original query. Indicates belonging to, Represents the original query set. This indicates that there exists only one. Represents a set of records. arg max means to return the record with the highest similarity. This represents the semantic similarity function.

[0078] 1. Experimental data: (1) Price database statistics The data in this experiment comes from actual engineering pricing scenarios to ensure the authenticity and practicality of the experiment. The price database serves as the target dataset for retrieval and contains 30,000 engineering pricing records, covering fields such as serial number, project code, project name, project characteristics, work content description, unit of measurement, 24-year labor guidance price, materials supplied by the client, main material loss rate, and remarks.

[0079] (2) Query set: The query set, based on a real-world pricing example, contains 500 search requests to simulate actual search scenarios. Fields include sequence number, project code, project name, project feature description, unit of measurement, labor cost, and brand. During the query process, matching items are retrieved from the price database based on the project name and feature description in the query set, which are then used to populate the project code and labor cost fields. The query set retains textual noise present in real-world applications (such as non-standard expressions), effectively validating the model's ability to handle complex inputs.

[0080] 2. Experimental Setup: In its implementation, this paper's method relies on existing open-source models and classic retrieval tools. During the query semantic enhancement stage, the large language model Qwen3-32B is employed, which possesses strong semantic understanding and instruction-following capabilities, effectively mitigating noise, abbreviations, and ambiguous expressions in the original query, thereby generating a more semantically standardized enhanced query. The dense semantic channel uses Qwen3-Embedding-8B as the text encoder, mapping queries and records to a shared vector space; the sparse term channel is implemented based on BM25, leveraging its advantages in matching explicit terms and numerical parameters. The scores of both channels are Min-Max normalized along the query-by-query dimension and then fused using linear weighting to form a hybrid channel. The validation set uses a fusion weight set to [value missing]. This achieves a better balance between semantic information and term matching. In verifying the impact of reranking on algorithm performance, the qwen3-reranker-0.6B model [Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models] was used.

[0081] 3. Evaluation Metrics To comprehensively evaluate the effectiveness of the proposed method in both candidate generation and ranking stages, this paper selects Recall@K and Mean Reciprocal Rank (MRR) as core metrics to evaluate system performance from the two dimensions of coverage and ranking quality, respectively.

[0082] First, Recall@K measures whether, during the candidate generation phase, the true target record can be included in a size of [size missing]. The candidate set is defined as follows:

[0083] Where N represents the number of test queries, r i For the i-th query, Represents the corresponding candidate set. This is an indicator function. A high Recall@K value indicates that the candidate generation module can effectively cover the correct answer.

[0084] Secondly Used to evaluate the overall quality of the ranking phase, particularly the relative position of the target record within the candidate set. It is defined as: , in, Represents the actual target record r i The position of the correct answer in the ranking results. A higher MRR value indicates that the model is able to more consistently rank the correct answer at the top, thus improving the final discrimination performance.

[0085] 4. Baselines To verify the effectiveness of the proposed model, four representative retrieval models were selected as baselines: (1)Qwen3-Embedding-8B [Qwen3 Embedding: Advancing Text Embedding andReranking Through Foundation Models] The Qwen3 series features the latest proprietary text embedding model with 8 billion parameters, supports 100+ natural languages ​​and multiple programming languages, and is widely used in retrieval, clustering, and ranking tasks.

[0086] (2)SimCSE [Fengshenbang 1.0: Being the Foundation of ChineseCognitive Intelligence] SimCSE is an unsupervised sentence vector representation model within a contrastive learning framework, capable of capturing sentence-level semantic similarity. The Erlangshen-SimCSE-110M-Chinese version we selected is optimized for Chinese, with a parameter size of 110 million, and is commonly used for Chinese retrieval and similarity calculation tasks.

[0087] (3)BM25 [The Probabilistic Relevance Framework: BM25 and Beyond] The classic probability-based sparse retrieval model takes into account factors such as word frequency, inverse document (record) frequency, and document (record) length normalization, and has robust performance in explicit keyword and parameter matching.

[0088] (4)TF-IDF [A statistical interpretation of term specificity and its application in retrieval] By combining term frequency (TF) and inverse document (record) frequency (IDF) to measure term importance, the discriminative power of sparse representations can be improved, making it one of the most commonly used benchmark retrieval methods.

[0089] 5. Experimental Results To comprehensively evaluate the effectiveness of the proposed method, experiments were conducted on a real engineering material price retrieval dataset, and the analysis was carried out from two aspects: overall performance and ablation comparison.

[0090] Table 1. Main Experiment Analysis

[0091] Table 1 shows a comparison of retrieval performance under different channel combinations. The results show that a single channel has limitations in both recall and ranking quality, while the dual-channel fusion scheme proposed in this invention achieves the best performance in both recall and MRR dimensions.

[0092] From the single-channel results, the dense semantic channel based on Qwen3-Embedding achieved recall rates of 84.76% and 88.37% for Recall@20 and Recall@30, respectively, indicating its ability to effectively capture semantic synonyms and contextual associations, demonstrating strong recall capabilities. However, its MRR index was relatively low, reflecting shortcomings in ranking accuracy. In contrast, the sparse term channel based on BM25 achieved recall rates of 78.12% and 81.16% for Recall@20 and Recall@30, respectively, significantly lower than the dense channel, but its MRR performance was similar, indicating that term matching still plays an important role in accurate ranking.

[0093] Overall, the dual-channel fusion scheme proposed in this invention achieves recall rates of 86.98% and 90.58% for Recall@20 and Recall@30, respectively, and MRR@20 and MRR@30 rates of 56.82% and 56.43%, respectively, all significantly outperforming single-channel methods. This result demonstrates that the fusion scheme can fully leverage the recall advantages of dense semantic modeling and the accuracy of sparse term matching, achieving effective complementarity between the two.

[0094] Table 2 Ablation Experiment

[0095] Further ablation experiments were conducted to verify the roles of each module in the overall algorithm. The experiments removed or combined the query semantic enhancement, dense semantic channel, sparse term channel, and re-ranking modules, and evaluated them on the same dataset. The results are shown in Table 2.

[0096] First, when using the dense semantic channel (Qwen3 Embedding) alone, Recall@20 reaches 84.76%, indicating that the semantic vector can effectively capture the deep semantic relevance between the query and the material description. However, its MRR@20 is only 50.86%, showing shortcomings in the fine-grained ranking of candidate results. In contrast, when using the sparse term channel (BM25) alone, Recall@20 is only 78.12%, significantly lower than embedding, but its MRR performance is close. This suggests that BM25 has certain advantages in the precise matching of explicit terms, but its overall coverage is limited.

[0097] Furthermore, when query semantic enhancement is introduced, the algorithm's Recall@20 increases from 82.83% to 86.98%, MRR@20 increases from 52.07% to 56.82%, and it also improves by approximately 4 percentage points in both Recall@30 and MRR@30. This indicates that query semantic enhancement can effectively alleviate the semantic drift problem by making key elements such as specification parameters and units explicit, thereby simultaneously improving candidate coverage and ranking accuracy.

[0098] It is worth noting that while introducing a reranking module into single-channel retrieval results can bring some improvement, it failed to provide further gains under the dual-channel fusion framework, and even declined in some metrics. This may be because the fusion strategy itself effectively takes into account both semantics and term matching, resulting in a more balanced candidate result. However, the general reranking model, lacking domain adaptation, weakens this balance. Based on this, the final method abandons the reranker and directly adopts the result of dual-channel fusion and dynamic weight learning.

[0099] Figure 2 Different weight parameters are shown Impact on model retrieval performance. The curve trend shows that, as... With the increase of , Recall@10, Recall@15, and Recall@20 all show a monotonically increasing trend, indicating that the semantic channel contributes significantly to the overall effect. At lower levels, the model relies excessively on sparse channels, resulting in significantly insufficient performance; when As the value gradually increases, the recall value rises rapidly, indicating that dense semantic representation has an advantage in capturing complex query semantics. However, when... After exceeding 0.7, the rate of increase of the curve gradually slowed down and stabilized in the range of 0.7 to 0.9, indicating that excessive reliance on semantic channels could no longer bring significant benefits.

[0100] Therefore, we can conclude that, on the one hand, semantic channels contribute more to the task than sparse channels, but complete reliance is still not robust enough; on the other hand, the fusion strategy works best with medium to high semantic weights, and its performance is relatively stable in the range of 0.7 to 0.9. Therefore, during the interval pruning process in the inference phase, we set the lower and upper bounds of the intervals to... .

[0101] This invention proposes an intelligent retrieval method for engineering material prices based on fine-grained field-aware fusion, effectively alleviating the shortcomings in retrieval accuracy and stability caused by low retrieval efficiency, long processing time, and subjective human intervention. This method achieves standardized expression of non-standardized queries through semantic alignment and structure-aware enhancement; simultaneously, it combines the synergistic fusion of dense semantic channels and sparse term channels, and introduces a dynamic weight learning mechanism to improve term matching accuracy while ensuring semantic understanding. Experimental results show that this method outperforms traditional baselines in metrics such as Recall@K and MRR, achieving a balance between candidate coverage and ranking accuracy.

[0102] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. An intelligent search method for engineering material prices based on fine-grained field-aware fusion, characterized in that, Includes the following steps: Step 1: Receive the unstructured raw query set and provide a semi-structured material library; Step 2: Perform semantic alignment and structure-aware enhancement processing on the original queries in the original query set to generate normalized enhanced queries; Step 3: Calculate the global similarity between the normalized enhanced query and each record in the material database using both dense semantic channels and sparse term channels; specifically including: Step 31: Construct a dense-sparse dual-channel framework, including a dense semantic channel and a sparse term channel; Step 32, in the dense semantic channel, a pre-trained semantic encoder is employed The normalized enhanced query and each record in the material library are mapped to a query semantic vector and a record vector respectively , and in, express Normalization operation; , Generally refers to a pre-trained semantic encoder The encoded result; Represents the real number field; Represents a set of d-dimensional real vectors; Query enhancement through normalization using the cosine inner product of vectors. and each record in the material library The similarity is used to obtain the global dense semantic similarity. : in, Represents the query semantic vector and record vector inner product operation; Step 33: In the sparse term channel, based on the word segmentation function The term set is obtained, and the normalized enhanced query is calculated using the BM25 model. With each record in the material library The term similarity between terms is used to obtain the global sparse term similarity. : in, Representing words, For words In Record word frequency in This represents a hyperparameter that adjusts word frequency saturation; its function is to control word frequency. The degree of influence on the final relevance score; Used to control the intensity of normalization; Representing records Length, This represents the average record length. The inverse recording frequency is defined as: in, To record the total, For words containing The number of records; Step 4: Perform field-level parsing on the normalized enhanced query and each record in the material library, and calculate the field-level fine-grained similarity of the dense semantic channel and the sparse term channel on the predefined key field set respectively; Step 5: Normalize the global similarity and field-level fine-grained similarity, obtain the global fusion weight and field-level fusion weight, perform weighted fusion, and generate the final fusion similarity. Step 6: Sort all records in the material library according to the final fusion similarity, and output the record with the highest ranking as the target material price information.

2. The intelligent retrieval method for engineering material prices based on fine-grained field-aware fusion as described in claim 1, characterized in that, Step 1 specifically includes: Step 11: Receive the unstructured raw query set for the engineering material price retrieval scenario. ,in, Indicates the first A raw query described in natural language. Indicates the index of the original query. This represents the total number of original queries, which include noise, industry abbreviations, and colloquial expressions. Step 12: Provide a semi-structured material library ,in, For structural patterns, For recording sets, This represents the j-th record, where j represents the index of the record, and n represents the total number of records. Each record... It consists of multiple heterogeneous fields, among which, This indicates the first heterogeneous field in the j-th record. This indicates the second heterogeneous field in the j-th record. This indicates the k-th heterogeneous field in the j-th record, where k represents the total number of heterogeneous fields; the heterogeneous fields include structured numeric / enumerated fields and free text fields.

3. The intelligent retrieval method for engineering material prices based on fine-grained field-aware fusion as described in claim 1, characterized in that, Step 2 specifically includes: Step 21: Perform semantic alignment processing on each original query in the original query set to generate semantically equivalent standardized queries; Step 22: Perform structure-aware enhancement processing on the standardized query to form a structured query field representation; Step 23: Concatenate the standardized query with the query field representation to generate a normalized enhanced query.

4. The intelligent retrieval method for engineering material prices based on fine-grained field-aware fusion as described in claim 3, characterized in that, Step 21 specifically includes: The semantic alignment process includes at least one of the following operations: synonym normalization, abbreviation expansion, and unit normalization. Through synonym normalization mapping, abbreviation expansion rule set, and unit normalization conversion function, noise, abbreviations, and fuzzy expressions in the original query are converted into standard forms, generating semantically equivalent standardized queries. ; Step 22 specifically includes: Step 221: Define the domain field set : Among them, the domain field set is divided into materials according to the engineering domain. ,Specification and unit ; Step 222: Using rule matching, dictionary mapping, or numerical recognition methods, extract data from standardized queries. Extract predefined key fields and their corresponding values ​​to obtain a structured query field representation. : in, Indicates key fields, Key fields The range of values, Key fields The value of ; Step 23 specifically includes: Standardized query With structured query field representation Concatenating and serializing results in normalized enhanced queries. : in, This is a concatenation function.

5. The intelligent retrieval method for engineering material prices based on fine-grained field-aware fusion as described in claim 4, characterized in that, The synonym normalization, abbreviation expansion, and unit normalization specifically include: Synonym normalization specifically involves: normalizing domain-specific vocabulary. abbreviations or aliases in Mapping to standard words Normalization yields vocabulary ; Abbreviation expansion utilizes rule sets. Replace the abbreviation 'a' with the standard form 'b'; Unit normalization is for numerical-unit pairs Using conversion functions This will be transformed into a unified measurement system.

6. The intelligent retrieval method for engineering material prices based on fine-grained field-aware fusion as described in claim 1, characterized in that, Step 4 specifically includes: Step 41: Parse the corresponding content of the same key field from each record in the normalized enhanced query and the material library respectively; Step 42: For each key field f, use the semantic encoder of the dense semantic channel to calculate the similarity between its field content vectors, and obtain the field-level dense similarity. ; Step 43: For each key field f, use the BM25 model of the sparse term channel to calculate the similarity between its field content terms, and obtain the field-level sparse term similarity. .

7. The intelligent retrieval method for engineering material prices based on fine-grained field-aware fusion as described in claim 1, characterized in that, Step 5 includes: Step 51: Apply the Min-Max normalization operator to the global dense semantic similarity. Global sparse term similarity Field-level dense similarity Similarity to field-level sparse terms Normalize it to unify its scale to the [0,1] interval: in, express Normalization results express Normalization results express Normalization results express Normalization results; Step 52: Input the semantic vector of the normalized enhanced query into the trained multilayer perceptron model to predict the global fusion weight and multiple field-level fusion weights for the current query; Step 53: Use the global fusion weight to linearly weight the normalized global dense semantic similarity and the normalized global sparse term similarity to obtain the dual-channel fusion similarity of global semantics. : in, For global fusion weights; Step 54: Apply the field-level fusion weights to linearly weight the normalized field-level dense similarity and normalized field-level sparse similarity of the corresponding fields to obtain the dual-channel fusion similarity of field-level semantics. : in, Assign weights to each field level; Step 55: Fuse the dual-channel similarity of global semantics. Dual-channel fusion similarity with various field-level semantics Summing and dividing by the number of fields plus one yields the final fused similarity. : in, The similarity is a dual-channel fusion of global semantics. For field-level semantic dual-channel fusion similarity, Indicates key fields, Represents a collection of domain fields.

8. The intelligent retrieval method for engineering material prices based on fine-grained field-aware fusion as described in claim 7, characterized in that, The training process of the multilayer perceptron model includes: 1) Constructing training samples: For the i-th normalized enhanced query In the discrete weight set Enumeration weight value Choose the option that allows the true target to be recorded. Enter based on Optimal weights of the retrieved Top-K ranking candidate set As a reference value; in, For indicator functions, Indicates the record being retrieved; Representation and Normalization Enhanced Queries The most similar One record; Similarly, for the i-th normalized enhanced query In the discrete weight set Enumeration weight value Choose the option that allows the true target to be recorded. Enter based on Optimal weights of the retrieved Top-K ranking candidate set As a reference value; Obtain the adaptive weight training sample set for: in, To query the training set; 2) Using the query semantic vector For input, a multilayer perceptron is used. Output fusion weights As a predicted value: in, The neural network representing the multilayer perceptron maps d-dimensional vectors to prediction weights; Represents a set of d-dimensional real vectors; Similarly, with the query semantic vector For input, a multilayer perceptron is used. Output field-level fusion weights As a predicted value: 3) During the training phase, the optimization objective is to minimize the mean squared error between the predicted and reference values, while introducing... The regularization term is used to train the multilayer perceptron model using the AdamW optimizer: in, These represent the learning parameters of the weighted scoring model. This represents the loss function that guides the training of a neural network. The regularization coefficient is . express Regular terms.

9. The intelligent retrieval method for engineering material prices based on fine-grained field-aware fusion as described in claim 8, characterized in that, During the inference phase, the fusion weights predicted by the multilayer perceptron model are... Perform range clipping operation: in, This means truncating the weights to a closed interval. Inside, The value is selected as [0.7, 0.9].

Images