Background quotation data-based feedback pushing method and system

Abstract

The invention provides a feedback pushing method and system based on background quotation data, and relates to the technical field of data processing, and the method comprises the steps: receiving a quotation request uploaded by a client, carrying out the semantic analysis, obtaining a commodity name keyword, carrying out the extension of the commodity name keyword, carrying out the standardization processing of a commodity attribute, and determining the feature of the commodity attribute; determining customer quotation preference features, and comprehensively obtaining multi-dimensional quotation request feature vectors; constructing a combined retrieval condition, extracting feature vectors of the target quotation data, performing weighting, performing screening through a multi-criterion decision model, generating a comprehensive matching degree matrix, and performing screening to obtain high-quality target quotation data; and sorting, determining optimal quotation data, pushing the optimal quotation data, receiving feedback information and feedback scores, if the feedback information is acceptance, generating corresponding order information and sending the order information to a supplier, and if the feedback information is rejection, pushing the optimal quotation data again.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a feedback push method and system based on background quotation data. Background Art

[0002] With the rapid development of e-commerce, online sales have become an important form of modern business activities. In the transaction process between enterprises and customers, quotation is a key link, and its quality and efficiency directly affect the success or failure of the transaction. The traditional quotation method usually relies on manual screening and push by sales staff, which has problems such as slow response speed, low personalization, and low conversion rate.

[0003] In addition, existing quotation push is often one-way, lacking the collection and use of customer feedback information. Customers' feedback behaviors such as acceptance, rejection, and evaluation of quotations imply their true preferences and opinions, which play an important guiding role in improving subsequent push. However, existing technologies rarely incorporate customer feedback into the closed loop of quotation push, resulting in the disconnection between pushed quotations and customer needs, making it difficult to continuously optimize.

[0004] Therefore, a solution is urgently needed to solve the problems existing in the existing quotation push. Summary of the invention

[0005] The embodiment of the present invention provides a feedback push method and system based on background quotation data, which can at least solve some problems existing in the prior art.

[0006] A first aspect of an embodiment of the present invention provides a feedback push method based on background quotation data, comprising:

[0007] Receive a quotation request uploaded by a client, wherein the quotation request includes a product name, product attributes and customer information, wherein the product attributes include product specifications, product materials and product quantities, perform semantic analysis on the product name, extract product name keywords through natural language processing technology, expand the product name keywords through preset synonym expansion rules to obtain a product name keyword set, perform standardization on the product attributes, map them through an attribute mapping matrix and determine product attribute characteristics by calculating similarity, determine customer quotation preference characteristics based on the customer information, fuse them in combination with customer portrait data, construct a customer comprehensive feature vector, and obtain a multi-dimensional quotation request feature vector based on the data obtained through comprehensive calculation;

[0008] Based on the multi-dimensional quotation request feature vector, a combined retrieval condition based on semantic similarity and attribute similarity is constructed, and the target quotation data matching the combined retrieval condition is retrieved in combination with the improved inverted index technology, and the feature vector of the target quotation data is extracted, including supplier characteristics, quotation characteristics, supply cycle characteristics and scoring characteristics, and the quotation characteristics and the supply cycle characteristics are weighted by adaptive weights, and the scoring characteristics are attenuated in combination with a preset time attenuation function, and the processed target quotation data is screened by a preset multi-criteria decision model, and a comprehensive matching degree matrix is ​​generated by calculation and screened to obtain high-quality target quotation data;

[0009] The high-quality target quotation data are sorted, the optimal quotation data is determined according to the sorting result, the optimal quotation data is pushed to the client based on the customer information, and the feedback information and feedback score of the client on the optimal quotation data are received. If the feedback information is to accept the optimal quotation data, the corresponding order information is generated and sent to the supplier related to the optimal quotation data. If the feedback information is to reject the optimal quotation data, the optimal quotation data is updated and pushed again.

[0010] In an optional embodiment,

[0011] Receive a quotation request uploaded by a client, wherein the quotation request includes a product name, product attributes and customer information, wherein the product attributes include product specifications, product materials and product quantities, perform semantic analysis on the product name, extract product name keywords through natural language processing technology, expand the product name keywords through preset synonym expansion rules, obtain a product name keyword set, standardize the product attributes, map them through an attribute mapping matrix and determine product attribute features by calculating similarity, determine customer quotation preference features based on the customer information, combine with customer portrait data for fusion, and construct a customer comprehensive feature vector. The data obtained by comprehensive calculation obtains a multi-dimensional quotation request feature vector including:

[0012] Receive a quotation request uploaded by a client, wherein the quotation request includes a product name, product attributes, and customer information, and the product attributes include product specifications, product materials, and product quantity;

[0013] For the product name, the product name is segmented and tagged with parts of speech by a word segmentation tool, and the key entities in the product name are identified in combination with a pre-trained named entity recognition model to generate an entity recognition result, and the keywords in the product name are extracted in combination with the part-of-speech tagging result to obtain an initial keyword set, and a synonym word list is constructed based on the initial keyword set through a synonym expansion mechanism, and for each element in the initial keyword set, a match is performed in the synonym word list to determine the corresponding synonym, and the matched synonyms are added to the initial keyword set to obtain the product name keyword set;

[0014] For each commodity type, a corresponding attribute mapping matrix is ​​constructed, wherein the rows of the attribute mapping matrix represent standardized attributes, the columns represent non-standardized attributes, and the elements in the attribute mapping matrix represent the mapping relationship between two commodity attributes. For the commodity attributes in the quotation request, the standardized attributes of each commodity attribute in the attribute mapping matrix are determined and the non-standardized attributes are converted into standardized attributes. The converted standardized attributes are converted into vector representations, and the commodity attribute features corresponding to the current commodity attributes are extracted through one-hot encoding;

[0015] Based on the customer information, according to the customer's historical quotation request data, determine the customer quotation preference characteristics corresponding to the current customer through an association rule mining algorithm, obtain customer portrait data including basic information, consumption behavior and credit rating from the customer information system, and fuse the customer quotation preference characteristics with the customer portrait data to obtain the customer comprehensive feature vector;

[0016] The product name keyword set, the product attribute features and the customer comprehensive feature vector are concatenated to obtain the multi-dimensional quotation request feature vector.

[0017] In an optional embodiment,

[0018] The customer quotation preference characteristics corresponding to the current customer are determined by the association rule mining algorithm as shown in the following formula:

[0019]

[0020] Where p represents the vector representation of customer quotation preference features, n represents the total number of rules obtained by the association rule mining algorithm, d represents a single quotation request in the customer historical quotation request dataset D, and X i represents the premise part of the i-th association rule, Y i represents the conclusion part of the i-th association rule, v i The binary vector representation of the preference attribute combination corresponding to the i-th rule.

[0021] In an optional embodiment,

[0022] Based on the multi-dimensional quotation request feature vector, a combined retrieval condition based on semantic similarity and attribute similarity is constructed, and the target quotation data matching the combined retrieval condition is retrieved in combination with the improved inverted index technology, and the feature vector of the target quotation data is extracted, including supplier characteristics, quotation characteristics, supply cycle characteristics and scoring characteristics, and the quotation characteristics and the supply cycle characteristics are weighted by adaptive weights, and the scoring characteristics are attenuated in combination with a preset time attenuation function, and the processed target quotation data is screened by a preset multi-criteria decision model, and a comprehensive matching degree matrix is ​​generated by calculation and screened to obtain high-quality target quotation data, including:

[0023] Based on the multi-dimensional quotation request feature vector, original quotation data corresponding to the quotation request is defined, the quotation request and the original quotation data are preprocessed to generate a normalized text representation, words in the normalized text representation are converted into word vectors through a preset word embedding model, a vector representation of the normalized text is obtained through weighted averaging, and a cosine similarity between the quotation request and the original quotation data is calculated and used as a semantic similarity;

[0024] Normalizing the structured attributes of the quotation request and the original quotation data and mapping them to the interval of 0 to 1, calculating the relative difference between the attribute value of the quotation request and the attribute value of the original quotation data for numerical attributes, determining whether the two attribute values ​​completely match for categorical attribute values, weighting and summing the similarity of each attribute to obtain a comprehensive attribute similarity, determining the comprehensive similarity through weighted summation based on the semantic similarity and the comprehensive attribute similarity, and using the comprehensive similarity as a measure of the combined search condition;

[0025] Based on the combined retrieval condition, the candidate quotation data is determined by an improved inverted index technology, the keyword clusters and attribute clusters of the candidate quotation data are determined by a clustering algorithm, a cluster identification code is assigned to each keyword cluster and attribute cluster and a corresponding inverted index is established, the keyword vector and attribute vector of the candidate quotation data are determined by word embedding and attribute normalization processing, the similarity between each keyword vector and the keyword cluster center vector and the similarity between each attribute vector and the attribute cluster center vector are calculated respectively, the top ten keyword clusters and the top ten attribute clusters with the highest similarity are selected and the corresponding candidate quotation data sets are obtained, the combined retrieval condition is calculated and sorted for the selected candidate quotation data, and the candidate quotation data with the highest comprehensive similarity is used as the target quotation data;

[0026] Feature extraction is performed on the target quotation data to obtain feature vectors corresponding to supplier features, quotation features, supply cycle features and scoring features, weighted processing is performed on the quotation features and the supply cycle features through an adaptive weight mechanism, the scoring features are decayed through a time decay function, the processed target quotation data are screened in combination with a pre-set multi-criteria decision model, the comprehensive matching degree matrix corresponding to each target quotation data is calculated, the evaluation score corresponding to each target quotation data is calculated based on the comprehensive matching degree matrix and compared with a preset evaluation score threshold, if the evaluation score is greater than the evaluation score threshold, the current target quotation data is retained as high-quality target quotation data, otherwise it is deleted, and multiple high-quality target quotation data are obtained.

[0027] In an optional embodiment,

[0028] Assign weights to each attribute as shown in the following formula:

[0029]

[0030] Among them, ω i represents the weight of the i-th attribute, λ1 represents the range weight, r i represents the value range of the i-th attribute, λ2 represents the stability weight, and s i represents the stability of the i-th attribute, λ3 represents the discrimination weight, and d i represents the distinguishability of the i-th attribute, r j Indicates the value range of the jth attribute, s j represents the stability of the jth attribute, d j represents the distinguishability of the jth attribute, and n represents the total number of attributes.

[0031] In an optional embodiment,

[0032] Sorting the high-quality target quotation data, determining the best quotation data according to the sorting result, pushing the best quotation data to the client based on the customer information, receiving the feedback information and feedback score of the best quotation data from the client, if the feedback information is to accept the best quotation data, generating corresponding order information and sending it to the supplier related to the best quotation data, if the feedback information is to reject the best quotation data, updating the best quotation data and re-pushing it includes:

[0033] Based on the high-quality target quotation data, the data are sorted by an improved approximate ideal solution sorting method to generate a decision matrix for the high-quality target quotation data, the decision matrix is ​​standardized to generate a standard matrix, a weighted standardized decision matrix is ​​calculated in combination with customer weight preferences, a positive ideal solution and a negative ideal solution are determined based on the weighted standardized decision matrix, a weighted Euclidean distance between each high-quality target quotation data and the positive ideal solution and the negative ideal solution is calculated, and a relative closeness coefficient is defined based on the weighted Euclidean distance;

[0034] Taking the positive ideal solution as a reference sequence and the high-quality target quotation data as a comparison sequence, calculating the correlation coefficient between the comparison sequence and the reference sequence, calculating the grey correlation degree based on the correlation coefficient, calculating the comprehensive evaluation value of each high-quality target quotation data based on the grey correlation degree and the relative closeness coefficient, sorting in descending order based on the comprehensive evaluation value, and taking the high-quality target quotation data ranked first as the optimal quotation data;

[0035] Based on the customer information, the best quotation data is sent to the client and the feedback information and feedback score of the best quotation data from the client are received. If the feedback information is the best quotation data received, the corresponding order information is generated and sent to the corresponding supplier for confirmation and execution, the current feedback information is added to the customer comprehensive feature vector and the comprehensive evaluation value is updated based on the feedback score corresponding to the current feedback information;

[0036] If the feedback information is to reject the optimal quotation data, a penalty item coefficient is defined and the penalty item of the comprehensive matching degree matrix is ​​adjusted and updated. The optimal quotation data is updated according to the adjusted comprehensive matching degree matrix and pushed to the client again.

[0037] In an optional embodiment,

[0038] The grey correlation degree is calculated based on the correlation coefficient as shown in the following formula:

[0039]

[0040] Among them, ψ m represents the grey correlation degree of the mth high-quality target quotation data, M represents the number of evaluation indicators, x 0n represents the value of the reference sequence on the nth evaluation index, x mn represents the value of the mth comparison series on the nth evaluation index, ρ represents the resolution coefficient, It represents the maximum absolute difference between all comparison series and reference series in each evaluation index. It represents the minimum absolute difference between all comparison series and reference series in each evaluation index.

[0041] A second aspect of an embodiment of the present invention provides a feedback push system based on background quotation data, including:

[0042] The first unit is used to receive a quotation request uploaded by a client, wherein the quotation request includes a product name, product attributes and customer information, the product attributes include product specifications, product materials and product quantities, perform semantic analysis on the product name, extract product name keywords through natural language processing technology, expand the product name keywords through preset synonym expansion rules to obtain a product name keyword set, perform standardization on the product attributes, map them through an attribute mapping matrix and determine product attribute characteristics by calculating similarity, determine customer quotation preference characteristics according to the customer information, combine with customer portrait data for fusion, construct a customer comprehensive feature vector, and obtain a multi-dimensional quotation request feature vector based on the data obtained through comprehensive calculation;

[0043] The second unit is used to construct a combined search condition based on semantic similarity and attribute similarity based on the multi-dimensional quotation request feature vector, retrieve target quotation data matching the combined search condition in combination with the improved inverted index technology, extract the feature vector of the target quotation data, including supplier characteristics, quotation characteristics, delivery cycle characteristics and scoring characteristics, weight the quotation characteristics and the delivery cycle characteristics by adaptive weights, perform attenuation processing on the scoring characteristics in combination with a preset time attenuation function, screen the processed target quotation data by a preset multi-criteria decision model, generate a comprehensive matching degree matrix by calculation, and obtain high-quality target quotation data by screening;

[0044] The third unit is used to sort the high-quality target quotation data, determine the best quotation data according to the sorting result, push the best quotation data to the client based on customer information, receive feedback information and feedback score from the client on the best quotation data, and if the feedback information is to accept the best quotation data, generate corresponding order information and send it to the supplier related to the best quotation data; if the feedback information is to reject the best quotation data, update the best quotation data and push it again.

[0045] According to a third aspect of the embodiments of the present invention,

[0046] An electronic device is provided, comprising:

[0047] processor;

[0048] a memory for storing processor-executable instructions;

[0049] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0050] A fourth aspect of the embodiments of the present invention is:

[0051] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0052] In the present invention, by performing semantic analysis on the commodity names, using natural language processing technology to extract keywords and perform synonym expansion, a more comprehensive and accurate commodity name table is obtained, and customer quotation preference characteristics and customer portrait data are comprehensively considered to construct a customer comprehensive feature vector, thereby improving the understanding and processing accuracy of quotation requests, and the processed target quotation data is screened through a preset multi-criteria decision model to generate a comprehensive matching degree matrix to obtain high-quality target quotation data, thereby achieving accurate quotation data retrieval and matching, and improving the quality and pertinence of quotation schemes. Through automated quotation request processing, data retrieval, matching, sorting and pushing processes, the quotation efficiency is greatly improved, and the workload of manual processing is reduced. In summary, the present invention achieves accurate quotation data retrieval and matching, optimizes the sorting and pushing processes of quotation data, improves quotation efficiency and quality, enhances data processing and analysis capabilities, and provides enterprises with intelligent quotation solutions with significant technical effects and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flow chart of a feedback push method based on background quotation data according to an embodiment of the present invention;

[0054] Figure 2 It is a structural diagram of a feedback push system based on background quotation data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0057] Figure 1 FIG. 1 is a flow chart of a method for pushing feedback based on background quotation data according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0058] S1. Receive a quotation request uploaded by a client, wherein the quotation request includes a product name, product attributes and customer information, wherein the product attributes include product specifications, product materials and product quantities, perform semantic analysis on the product name, extract product name keywords through natural language processing technology, expand the product name keywords through preset synonym expansion rules to obtain a product name keyword set, perform standardization on the product attributes, map them through an attribute mapping matrix and determine product attribute features by calculating similarity, determine customer quotation preference features based on the customer information, combine with customer portrait data for fusion, construct a customer comprehensive feature vector, and obtain a multi-dimensional quotation request feature vector based on the data obtained through comprehensive calculation;

[0059] The natural language processing technology is a branch of artificial intelligence, involving the interaction between machine and human language. The synonym expansion rules are used to improve the system's understanding and query capabilities in natural language processing and information retrieval, including dictionary expansion, context expansion and semantic network. The attribute mapping matrix is ​​a matrix used to represent the relationship between different entity or object attributes. The customer quotation preference feature describes the tendency characteristics shown by the customer when accepting or rejecting the quotation. The customer comprehensive feature vector is a multidimensional vector formed by integrating various characteristics of the customer (such as behavioral data, preference characteristics, demographic data, etc.).

[0060] In an optional embodiment,

[0061] Receive a quotation request uploaded by a client, wherein the quotation request includes a product name, product attributes and customer information, wherein the product attributes include product specifications, product materials and product quantities, perform semantic analysis on the product name, extract product name keywords through natural language processing technology, expand the product name keywords through preset synonym expansion rules, obtain a product name keyword set, standardize the product attributes, map them through an attribute mapping matrix and determine product attribute features by calculating similarity, determine customer quotation preference features based on the customer information, combine with customer portrait data for fusion, and construct a customer comprehensive feature vector. The data obtained by comprehensive calculation obtains a multi-dimensional quotation request feature vector including:

[0062] Receive a quotation request uploaded by a client, wherein the quotation request includes a product name, product attributes, and customer information, and the product attributes include product specifications, product materials, and product quantity;

[0063] For the product name, the product name is segmented and tagged with parts of speech by a word segmentation tool, and the key entities in the product name are identified in combination with a pre-trained named entity recognition model to generate an entity recognition result, and the keywords in the product name are extracted in combination with the part-of-speech tagging result to obtain an initial keyword set, and a synonym word list is constructed based on the initial keyword set through a synonym expansion mechanism, and for each element in the initial keyword set, a match is performed in the synonym word list to determine the corresponding synonym, and the matched synonyms are added to the initial keyword set to obtain the product name keyword set;

[0064] For each commodity type, a corresponding attribute mapping matrix is ​​constructed, wherein the rows of the attribute mapping matrix represent standardized attributes, the columns represent non-standardized attributes, and the elements in the attribute mapping matrix represent the mapping relationship between two commodity attributes. For the commodity attributes in the quotation request, the standardized attributes of each commodity attribute in the attribute mapping matrix are determined and the non-standardized attributes are converted into standardized attributes. The converted standardized attributes are converted into vector representations, and the commodity attribute features corresponding to the current commodity attributes are extracted through one-hot encoding;

[0065] Based on the customer information, according to the customer's historical quotation request data, determine the customer quotation preference characteristics corresponding to the current customer through an association rule mining algorithm, obtain customer portrait data including basic information, consumption behavior and credit rating from the customer information system, and fuse the customer quotation preference characteristics with the customer portrait data to obtain the customer comprehensive feature vector;

[0066] The product name keyword set, the product attribute features and the customer comprehensive feature vector are concatenated to obtain the multi-dimensional quotation request feature vector.

[0067] The word segmentation tool is a basic tool in natural language processing, which is used to divide continuous text into separate words or phrases. The named entity recognition model is a machine learning or deep learning model used to identify entities in text (such as names of people, places, organization names, dates, etc.). The synonym vocabulary is a dictionary containing words and their synonyms, which is used to expand queries and text understanding. The one-hot encoding is a method of converting categorical variables into binary vectors. The association rule mining algorithm is used to discover interesting relationships or association rules between items in a data set.

[0068] The client uploads the quotation request to the server through the network interface. After receiving the quotation request, the server parses the quotation request data, obtains the product name, product attributes and customer information contained therein, and further parses the product attributes into specific attributes such as product specifications, product materials and product quantity;

[0069] Use a word segmentation tool to perform word segmentation and part-of-speech tagging on the product name, split the product name into several words, and mark the part of speech of each word. Use a pre-trained named entity recognition model to identify key entities in the product name, identify key information such as brand, model, and specification contained in the product name, generate entity recognition results, combine the part-of-speech tagging results and entity recognition results, extract keywords from the product name, and build an initial keyword set. Combine the part-of-speech tagging results and entity recognition results to extract keywords from the product name and build an initial keyword set. For each keyword in the initial keyword set, match it in a pre-built synonym word forest, add the matched synonyms to the initial keyword set, expand the coverage of the keywords, and use the expanded keyword set as the final product name keyword set;

[0070] For each product category, an attribute mapping matrix is ​​pre-built. The rows of the attribute mapping matrix represent standardized product attributes, and the columns represent non-standardized product attributes. The elements in the matrix represent the mapping relationship between the two attributes, which is used to convert non-standardized attributes into standardized attributes. For each product attribute in the quotation request, a match is made in the corresponding attribute mapping matrix to determine the standardized attribute corresponding to the product attribute in the matrix, and the non-standardized product attributes are converted into standardized attributes. The standardized product attributes are converted into vector representations. For each standardized attribute, one-hot encoding is used for encoding, and the encoded attribute vectors are concatenated to obtain the product attribute feature vector.

[0071] Convert the customer's historical quotation request data into a transaction data set, use association rule mining algorithms such as Apriori and FP-Growth to discover frequent patterns and association rules in customer quotation behavior, summarize the customer's quotation preference characteristics based on the mined association rules, and obtain customer portrait data from the customer information system, including basic customer information such as age, gender, region, etc., customer consumption behavior data such as purchase frequency, purchase amount, purchase category, etc., customer credit rating data, reflecting the customer's credit status and performance ability, merge customer quotation preference characteristics and customer portrait data, perform feature engineering on customer portrait data, extract key features, and splice customer quotation preference characteristics and customer portrait features to obtain a comprehensive customer feature vector;

[0072] Convert the product name keyword set into a vector representation. You can use methods such as Bag-of-Words and TF-IDF to concatenate the product attribute feature vector with the product name keyword vector to obtain the product dimension feature vector. Concatenate the customer comprehensive feature vector with the product dimension feature vector to obtain a multi-dimensional quotation request feature vector.

[0073] In this embodiment, by deeply mining and integrating multi-dimensional data such as product names, product attributes and customer information, a multi-dimensional feature vector that comprehensively describes the quotation request is generated, which can reflect the characteristics and needs of the quotation request from multiple angles, and provide rich decision-making basis for subsequent quotation optimization and recommendation. By performing word segmentation, part-of-speech tagging and named entity recognition on the product name, key information in the product name is extracted, and the semantic expression of keywords is enriched through synonym expansion, which can more accurately understand the customer's quotation needs and improve the accuracy of the quotation. Through automated data processing and feature extraction, the original quotation request is converted into a structured multi-dimensional feature vector, which greatly reduces the workload of manual processing, can significantly improve processing efficiency, reduce labor costs, and realize the automation and intelligence of the quotation process. In summary, this embodiment improves the accuracy and personalization of the quotation, reduces the manual processing cost, enhances the scalability and application flexibility of the system, and plays an important role in promoting the intelligence and optimization of the entire quotation process.

[0074] In an optional embodiment,

[0075] The customer quotation preference characteristics corresponding to the current customer are determined by the association rule mining algorithm as shown in the following formula:

[0076]

[0077] Where p represents the vector representation of customer quotation preference features, n represents the total number of rules obtained by the association rule mining algorithm, d represents a single quotation request in the customer historical quotation request dataset D, and X i represents the premise part of the i-th association rule, Y i represents the conclusion part of the i-th association rule, v i The binary vector representation of the preference attribute combination corresponding to the i-th rule.

[0078] In this embodiment, by performing association rule mining on the customer's historical quotation request data, the frequent patterns and association rules of the customer in the quotation behavior can be discovered, which helps to deeply understand the customer's quotation decision-making mechanism and provide a basis for subsequent personalized quotation recommendations. The vector representation of the customer's quotation preference characteristics is calculated through a formula, which realizes the quantification and numericalization of the customer's preferences. It not only considers the frequency of the rules, but also reflects the confidence level of the rules, and can more accurately characterize the customer's preference intensity and priority. Through automated association rule mining and preference feature extraction, large-scale customer quotation data can be quickly processed to discover hidden preference patterns, which not only improves the efficiency of quotation, but also improves the quality and accuracy of quotation. In summary, this embodiment can deeply mine the customer's preference pattern, improve the personalization of quotation and recommendation quality, and at the same time improve quotation efficiency and expand application scenarios.

[0079] S2. Based on the multi-dimensional quotation request feature vector, a combined search condition based on semantic similarity and attribute similarity is constructed, and the target quotation data matching the combined search condition is retrieved in combination with the improved inverted index technology, and the feature vector of the target quotation data is extracted, including supplier characteristics, quotation characteristics, supply cycle characteristics and scoring characteristics, and the quotation characteristics and the supply cycle characteristics are weighted by adaptive weights, and the scoring characteristics are attenuated in combination with a preset time attenuation function, and the processed target quotation data is screened by a preset multi-criteria decision model, and a comprehensive matching degree matrix is ​​generated by calculation and screened to obtain high-quality target quotation data;

[0080] The combined retrieval condition refers to the use of multiple conditions for screening during the search or query process to improve the accuracy and relevance of the retrieval results. The inverted index is an efficient index data structure used to quickly find a document collection containing specific words. The time decay function is used to indicate that the importance of data gradually decreases over time in data analysis or information retrieval. The multi-criteria decision model is a method for making the best choice among multiple alternatives. The comprehensive matching degree matrix is ​​a data structure in the multi-criteria decision model, which is used to represent the comprehensive score of each alternative on each evaluation indicator.

[0081] In an optional embodiment,

[0082] Based on the multi-dimensional quotation request feature vector, a combined retrieval condition based on semantic similarity and attribute similarity is constructed, and the target quotation data matching the combined retrieval condition is retrieved in combination with the improved inverted index technology, and the feature vector of the target quotation data is extracted, including supplier characteristics, quotation characteristics, supply cycle characteristics and scoring characteristics, and the quotation characteristics and the supply cycle characteristics are weighted by adaptive weights, and the scoring characteristics are attenuated in combination with a preset time attenuation function. The processed target quotation data is screened through a preset multi-criteria decision model, and a comprehensive matching degree matrix is ​​generated by calculation and screened to obtain high-quality target quotation data, including:

[0083] Based on the multi-dimensional quotation request feature vector, original quotation data corresponding to the quotation request is defined, the quotation request and the original quotation data are preprocessed to generate a normalized text representation, words in the normalized text representation are converted into word vectors through a preset word embedding model, a vector representation of the normalized text is obtained through weighted averaging, and a cosine similarity between the quotation request and the original quotation data is calculated and used as a semantic similarity;

[0084] Normalizing the structured attributes of the quotation request and the original quotation data and mapping them to the interval of 0 to 1, calculating the relative difference between the attribute value of the quotation request and the attribute value of the original quotation data for numerical attributes, determining whether the two attribute values ​​completely match for categorical attribute values, weighting and summing the similarity of each attribute to obtain a comprehensive attribute similarity, determining the comprehensive similarity through weighted summation based on the semantic similarity and the comprehensive attribute similarity, and using the comprehensive similarity as a measure of the combined search condition;

[0085] Based on the combined retrieval condition, the candidate quotation data is determined by an improved inverted index technology, the keyword clusters and attribute clusters of the candidate quotation data are determined by a clustering algorithm, a cluster identification code is assigned to each keyword cluster and attribute cluster and a corresponding inverted index is established, the keyword vector and attribute vector of the candidate quotation data are determined by word embedding and attribute normalization processing, the similarity between each keyword vector and the keyword cluster center vector and the similarity between each attribute vector and the attribute cluster center vector are calculated respectively, the top ten keyword clusters and the top ten attribute clusters with the highest similarity are selected and the corresponding candidate quotation data sets are obtained, the combined retrieval condition is calculated and sorted for the selected candidate quotation data, and the candidate quotation data with the highest comprehensive similarity is used as the target quotation data;

[0086] Feature extraction is performed on the target quotation data to obtain feature vectors corresponding to supplier features, quotation features, supply cycle features and scoring features, weighted processing is performed on the quotation features and the supply cycle features through an adaptive weight mechanism, the scoring features are decayed through a time decay function, the processed target quotation data are screened in combination with a pre-set multi-criteria decision model, the comprehensive matching degree matrix corresponding to each target quotation data is calculated, the evaluation score corresponding to each target quotation data is calculated based on the comprehensive matching degree matrix and compared with a preset evaluation score threshold, if the evaluation score is greater than the evaluation score threshold, the current target quotation data is retained as high-quality target quotation data, otherwise it is deleted, and multiple high-quality target quotation data are obtained.

[0087] The raw quotation data refers to unprocessed quotation information directly obtained from suppliers or systems, usually including product name, specifications, price, supplier information, delivery time, etc. The word embedding model is a technology that maps words into low-dimensional vectors in a continuous vector space. The word vector is the result generated by the word embedding model and is a low-dimensional vector representing a word. The cosine similarity is a method for measuring the similarity between two vectors. The numerical attribute refers to an attribute that can be expressed by a numerical value, such as price, quantity, weight, etc. The categorical attribute refers to an attribute that represents a category or classification, such as product type, color, brand, etc. The keyword cluster is a set formed by clustering keywords with similar semantics together, and the attribute cluster is a set formed by clustering attributes with similar characteristics together.

[0088] Define the original quotation data corresponding to the quotation request, pre-process the quotation request and the original quotation data, generate a standardized text representation, perform word segmentation, stop word removal, part-of-speech tagging and other processing on the text to obtain a standardized word sequence, normalize the numerical attributes, map the attribute values ​​to a unified scale, use a pre-trained word embedding model to convert the words in the standardized text representation into word vectors, query the corresponding word vector in the word embedding model for each word, form a word vector sequence with the queried word vectors, merge the word vector sequence into a vector representation of the standardized text by weighted averaging, assign a weight to each word vector in the word vector sequence, average the weighted word vectors to obtain a vector representation of the standardized text, and calculate the cosine similarity between the standardized text vector representations of the quotation request and the original quotation data as the semantic similarity;

[0089] Normalize the structured attributes of the quotation request and the original quotation data, perform minimum-maximum normalization on the numerical attributes, map the attribute values ​​to the [0, 1] interval, perform one-hot encoding on the categorical attributes, and convert the attribute values ​​into binary vectors. For each attribute, calculate the attribute similarity between the quotation request and the original quotation data. For numerical attributes, calculate the relative difference between the attribute value of the quotation request and the attribute value of the original quotation data. For categorical attributes, determine whether the two attribute values ​​completely match. If they match, the similarity is 1, and if they do not match, it is 0. Multiply the similarity of each attribute by the corresponding weight and perform weighted summation to obtain the comprehensive attribute similarity.

[0090] Based on the combined retrieval conditions (including semantic similarity and comprehensive attribute similarity), the candidate quotation data is determined by the improved inverted index technology, the candidate quotation data is clustered, the keyword clusters and attribute clusters are determined by the clustering algorithm, a unique cluster identification code is assigned to each keyword cluster and attribute cluster, an inverted index corresponding to the keyword cluster and attribute cluster is established, the candidate quotation data is mapped to the corresponding cluster, the candidate quotation data is vectorized, the keywords of the candidate quotation data are converted into keyword vectors by using the word embedding model, the attributes of the candidate quotation data are normalized to obtain the attribute vector, the similarity between the candidate quotation data and the keyword cluster and the attribute cluster is calculated, the cosine similarity between each keyword vector and the keyword cluster center vector is calculated, the cosine similarity between each attribute vector and the attribute cluster center vector is calculated, the top ten keyword clusters and attribute clusters with the highest similarity are selected, and the corresponding candidate quotation data set is obtained, the combined retrieval conditions are calculated and sorted for the selected candidate quotation data, and the candidate quotation data with the highest comprehensive similarity is used as the target quotation data;

[0091] The target quotation data is feature extracted, the quotation features and the supply cycle features are weighted, and the weights of the quotation features and the supply cycle features are dynamically adjusted using an adaptive weight mechanism according to the importance and urgency of the quotation request. The weighted quotation features and the supply cycle features are normalized, the scoring features are time-decayed, and a time-decay function is designed to decay the historical scores, with the recent scores having a higher weight. The decayed scoring features are normalized, and the target quotation data are screened using a pre-set multi-criteria decision model. The processed supplier features, quotation features, supply cycle features, and scoring features are used as decision criteria, and the scores of each target quotation data under each criterion are calculated to form a comprehensive matching matrix. Based on the comprehensive matching matrix, the evaluation score of each target quotation data is calculated, and the evaluation score is compared with a preset evaluation score threshold. If the evaluation score is greater than the threshold, the current target quotation data is retained as high-quality target quotation data. If the evaluation score is less than or equal to the threshold, the current target quotation data is deleted, and multiple high-quality target quotation data after screening are output.

[0092] In this embodiment, by preprocessing the quotation request and the original quotation data and calculating the semantic similarity and the comprehensive attribute similarity, the quotation data can be matched from the two dimensions of semantics and attributes, and the semantic association and attribute similarity between the quotation request and the quotation data can be effectively captured to improve the matching accuracy. Through the improved inverted index technology, the candidate quotation data related to the quotation request can be quickly retrieved to further improve the matching efficiency. The time decay function is introduced to decay the historical scores so that the recent scores have a higher weight, which can reflect the dynamic performance changes of the supplier and ensure that the screened high-quality quotation data has high timeliness. In summary, this embodiment improves the matching accuracy and screening quality of the quotation data, improves the timeliness and dynamic adaptability of the quotation data, provides a solid data foundation for subsequent quotation optimization and decision-making, and improves the intelligence level and efficiency of the quotation business.

[0093] In an optional embodiment,

[0094] Assign weights to each attribute as shown in the following formula:

[0095]

[0096] Among them, ω i represents the weight of the i-th attribute, λ1 represents the range weight, r i represents the value range of the i-th attribute, λ2 represents the stability weight, and s i represents the stability of the i-th attribute, λ3 represents the discrimination weight, and d i represents the distinguishability of the i-th attribute, r j Indicates the value range of the jth attribute, s j represents the stability of the jth attribute, d j represents the distinguishability of the jth attribute, and n represents the total number of attributes.

[0097] In this embodiment, the three characteristics of the attribute value range, stability and distinguishability are comprehensively considered, and the importance and influence of each attribute can be evaluated more comprehensively and objectively. By adjusting the weight parameters, the calculation of the attribute weight can be flexibly controlled to adapt to different business needs and preferences. The adaptive adjustment mechanism enables the attribute weight to adapt to changes in data, improves the robustness of the weight calculation, and the weighted calculation method can more accurately reflect the similarity between attributes and improve the accuracy of attribute similarity calculation. In summary, this embodiment can improve the accuracy of attribute similarity calculation, enhance the effect of data analysis and mining, and provide the explainability and interpretability of weight calculation, which is helpful to extract key attributes, reduce noise interference, and provide a more reliable basis for subsequent data processing and decision-making.

[0098] S3. Sort the high-quality target quotation data, determine the best quotation data according to the sorting results, push the best quotation data to the client based on the customer information, receive the client's feedback information and feedback score on the best quotation data, and if the feedback information is to accept the best quotation data, generate corresponding order information and send it to the supplier related to the best quotation data; if the feedback information is to reject the best quotation data, update the best quotation data and push it again.

[0099] The optimal quotation data refers to the best quotation selected from multiple suppliers' quotations through a series of screening, sorting and optimization algorithms. The feedback information refers to the opinions and suggestions provided by the customer after receiving the quotation. The feedback score refers to the score given by the customer to various indicators after receiving the quotation.

[0100] In an optional embodiment,

[0101] Sorting the high-quality target quotation data, determining the best quotation data according to the sorting result, pushing the best quotation data to the client based on the customer information, receiving the feedback information and feedback score of the best quotation data from the client, if the feedback information is to accept the best quotation data, generating corresponding order information and sending it to the supplier related to the best quotation data, if the feedback information is to reject the best quotation data, updating the best quotation data and re-pushing it includes:

[0102] Based on the high-quality target quotation data, the data are sorted by an improved approximate ideal solution sorting method to generate a decision matrix for the high-quality target quotation data, the decision matrix is ​​standardized to generate a standard matrix, a weighted standardized decision matrix is ​​calculated in combination with customer weight preferences, a positive ideal solution and a negative ideal solution are determined based on the weighted standardized decision matrix, a weighted Euclidean distance between each high-quality target quotation data and the positive ideal solution and the negative ideal solution is calculated, and a relative closeness coefficient is defined based on the weighted Euclidean distance;

[0103] Taking the positive ideal solution as a reference sequence and the high-quality target quotation data as a comparison sequence, calculating the correlation coefficient between the comparison sequence and the reference sequence, calculating the grey correlation degree based on the correlation coefficient, calculating the comprehensive evaluation value of each high-quality target quotation data based on the grey correlation degree and the relative closeness coefficient, sorting in descending order based on the comprehensive evaluation value, and taking the high-quality target quotation data ranked first as the optimal quotation data;

[0104] Based on the customer information, the best quotation data is sent to the client and the feedback information and feedback score of the best quotation data from the client are received. If the feedback information is the best quotation data received, the corresponding order information is generated and sent to the corresponding supplier for confirmation and execution, the current feedback information is added to the customer comprehensive feature vector and the comprehensive evaluation value is updated based on the feedback score corresponding to the current feedback information;

[0105] If the feedback information is to reject the optimal quotation data, a penalty item coefficient is defined and the penalty item of the comprehensive matching degree matrix is ​​adjusted and updated. The optimal quotation data is updated according to the adjusted comprehensive matching degree matrix and pushed to the client again.

[0106] The approximate ideal solution sorting method is a multi-criteria decision analysis method. The core idea is to select a solution so that its distance from the ideal solution is the smallest and its distance from the negative ideal solution is the largest. The decision matrix is ​​a data table that describes the performance of each solution on different evaluation indicators. The positive ideal solution refers to the solution that reaches the optimal value on all evaluation indicators, and the negative ideal solution refers to the solution that reaches the worst value on all evaluation indicators. The weighted Euclidean distance takes into account the weights of each evaluation indicator when calculating the distance between the solution and the positive ideal solution or the negative ideal solution. The relative closeness coefficient is used to measure the closeness of the solution to the ideal solution. The gray correlation degree is used to measure the correlation between different solutions and the ideal solution. The penalty item coefficient is used to dynamically adjust the priority of the quotation solution that is not accepted by the customer to improve the matching degree of the subsequent pushed quotations.

[0107] Generate a decision matrix for high-quality target quotation data. The elements in the matrix represent the attribute values ​​of each quotation data under different evaluation indicators. Standardize the decision matrix, convert the attribute values ​​into dimensionless standardized values, generate a standard matrix, and calculate the weighted standardized decision matrix in combination with the customer's weight preference. Multiply the standardized value of each attribute by the corresponding weight coefficient. Based on the weighted standardized decision matrix, determine the positive ideal solution and the negative ideal solution. The positive ideal solution represents the optimal value of each evaluation indicator, and the negative ideal solution represents the worst value of each evaluation indicator. Calculate the weighted Euclidean distance between each high-quality target quotation data and the positive ideal solution and the negative ideal solution. Based on the weighted Euclidean distance, define the relative closeness coefficient , represents the closeness between the quotation data and the positive ideal solution, takes the positive ideal solution as the reference series, and the high-quality target quotation data as the comparison series, calculates the grey correlation, calculates the correlation coefficient between the comparison series and the reference series, represents the correlation between the two series at different times, and calculates the grey correlation based on the correlation coefficient, represents the overall correlation between the comparison series and the reference series, comprehensively considers the relative closeness coefficient and the grey correlation, calculates the comprehensive evaluation value of each high-quality target quotation data, performs a weighted combination of the relative closeness coefficient and the grey correlation, obtains the comprehensive evaluation value, sorts the high-quality target quotation data in descending order based on the comprehensive evaluation value, and takes the first place in the sorting result as the optimal quotation data;

[0108] Based on the customer information, the best quotation data is sent to the client, the feedback information and feedback score of the best quotation data from the client are received, and the type of feedback information is determined. If the feedback information is the best quotation data received, the corresponding order information is generated and sent to the corresponding supplier for confirmation and execution. The current feedback information is added to the customer comprehensive feature vector to update the customer preference. Based on the feedback score corresponding to the current feedback information, the comprehensive evaluation value of the best quotation data is updated;

[0109] If the feedback information is to reject the best quotation data, a penalty item coefficient is defined to adjust the comprehensive matching matrix, adjust and update the penalty item of the comprehensive matching matrix, reduce the matching degree of the rejected quotation data, recalculate the best quotation data according to the adjusted comprehensive matching matrix, and push the updated best quotation data to the client again, and repeat the push until the client accepts the best quotation data or the maximum number of iterations is reached.

[0110] In this embodiment, by constructing a decision matrix and performing standardization and weighting processing, the influence of multiple evaluation indicators on quotation data is taken into account, and the relative closeness coefficient and gray correlation are comprehensively considered to obtain a comprehensive evaluation value, so that the sorting result is more reliable and robust. By introducing customer preferences, the generation of quotation schemes is closer to customer expectations, which improves the pertinence of quotation schemes and customer satisfaction. The dynamic optimization mechanism can continuously improve the quality of quotation schemes and improve customer satisfaction. This automated quotation process shortens the quotation cycle and improves the timeliness and accuracy of quotations. In summary, this embodiment improves the scientificity, pertinence and dynamic optimization capabilities of quotation schemes, enhances quotation efficiency and automation, and has good explainability, traceability and large-scale data processing capabilities, providing enterprises with high-quality quotation decision support and improving the intelligence level of quotation business and customer satisfaction.

[0111] In an optional embodiment,

[0112] The grey correlation degree is calculated based on the correlation coefficient as shown in the following formula:

[0113]

[0114] Among them, ψ m represents the grey correlation degree of the mth high-quality target quotation data, M represents the number of evaluation indicators, x 0n represents the value of the reference sequence on the nth evaluation index, x mn represents the value of the mth comparison series on the nth evaluation index, ρ represents the resolution coefficient, It represents the maximum absolute difference between all comparison series and reference series in each evaluation index. It represents the minimum absolute difference between all comparison series and reference series in each evaluation index.

[0115] In this embodiment, by calculating the difference between the quotation data and the ideal solution in each evaluation index, and comprehensively considering the maximum and minimum values ​​of the difference, a normalized correlation value is obtained. The size of the gray correlation reflects the closeness between the quotation data and the ideal solution. The larger the correlation, the more similar the quotation data is to the ideal solution and the closer it is to the optimal solution. By averaging the correlation coefficient, the influence of different evaluation indicators can be balanced, the excessive dominance of a certain indicator can be avoided, and the comprehensiveness and reliability of the evaluation results can be improved. The relative evaluation method eliminates the influence of the dimensions of different evaluation indicators, so that the value range of the correlation is between [0, 1], which is convenient for comparison and sorting between different quotation data. The maximum and minimum values ​​of the difference between all quotation data and the ideal solution in each evaluation index are considered, and a global perspective is introduced to reduce the influence of individual outliers or extreme values ​​on the correlation, and improve the robustness of the evaluation results. In summary, this embodiment considers the comprehensive influence of multiple evaluation indicators, introduces a resolution coefficient to adjust the sensitivity of the correlation, realizes the relative evaluation of quotation data, improves the robustness of the evaluation results, and provides an important reference for selecting the optimal quotation scheme.

[0116] Figure 2 FIG. 1 is a schematic diagram of a feedback push system based on background quotation data according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0117] The first unit is used to receive a quotation request uploaded by a client, wherein the quotation request includes a product name, product attributes and customer information, the product attributes include product specifications, product materials and product quantities, perform semantic analysis on the product name, extract product name keywords through natural language processing technology, expand the product name keywords through preset synonym expansion rules to obtain a product name keyword set, perform standardization on the product attributes, map them through an attribute mapping matrix and determine product attribute characteristics by calculating similarity, determine customer quotation preference characteristics according to the customer information, combine with customer portrait data for fusion, construct a customer comprehensive feature vector, and obtain a multi-dimensional quotation request feature vector based on the data obtained through comprehensive calculation;

[0118] The second unit is used to construct a combined search condition based on semantic similarity and attribute similarity based on the multi-dimensional quotation request feature vector, retrieve target quotation data matching the combined search condition in combination with the improved inverted index technology, extract the feature vector of the target quotation data, including supplier characteristics, quotation characteristics, delivery cycle characteristics and scoring characteristics, weight the quotation characteristics and the delivery cycle characteristics by adaptive weights, perform attenuation processing on the scoring characteristics in combination with a preset time attenuation function, screen the processed target quotation data by a preset multi-criteria decision model, generate a comprehensive matching degree matrix by calculation, and obtain high-quality target quotation data by screening;

[0119] The third unit is used to sort the high-quality target quotation data, determine the best quotation data according to the sorting result, push the best quotation data to the client based on customer information, receive feedback information and feedback score from the client on the best quotation data, and if the feedback information is to accept the best quotation data, generate corresponding order information and send it to the supplier related to the best quotation data; if the feedback information is to reject the best quotation data, update the best quotation data and push it again.

[0120] According to a third aspect of the embodiments of the present invention,

[0121] An electronic device is provided, comprising:

[0122] processor;

[0123] a memory for storing processor-executable instructions;

[0124] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0125] A fourth aspect of the embodiments of the present invention is:

[0126] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0127] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. The feedback push method based on background quotation data is characterized by: include: Receive a quotation request uploaded by a client, wherein the quotation request includes a product name, product attributes and customer information, wherein the product attributes include product specifications, product materials and product quantities, perform semantic analysis on the product name, extract product name keywords through natural language processing technology, expand the product name keywords through preset synonym expansion rules to obtain a product name keyword set, perform standardization on the product attributes, map them through an attribute mapping matrix and determine product attribute characteristics by calculating similarity, determine customer quotation preference characteristics based on the customer information, fuse them in combination with customer portrait data, construct a customer comprehensive feature vector, and obtain a multi-dimensional quotation request feature vector based on the data obtained through comprehensive calculation; Based on the multi-dimensional quotation request feature vector, a combined retrieval condition based on semantic similarity and attribute similarity is constructed, and the target quotation data matching the combined retrieval condition is retrieved in combination with the improved inverted index technology, and the feature vector of the target quotation data is extracted, including supplier characteristics, quotation characteristics, supply cycle characteristics and scoring characteristics, and the quotation characteristics and the supply cycle characteristics are weighted by adaptive weights, and the scoring characteristics are attenuated in combination with a preset time attenuation function, and the processed target quotation data is screened by a preset multi-criteria decision model, and a comprehensive matching degree matrix is ​​generated by calculation and screened to obtain high-quality target quotation data; The high-quality target quotation data are sorted, the optimal quotation data is determined according to the sorting result, the optimal quotation data is pushed to the client based on the customer information, and the feedback information and feedback score of the client on the optimal quotation data are received. If the feedback information is to accept the optimal quotation data, the corresponding order information is generated and sent to the supplier related to the optimal quotation data. If the feedback information is to reject the optimal quotation data, the optimal quotation data is updated and pushed again.

2. The method according to claim 1, characterized in that Receive a quotation request uploaded by a client, wherein the quotation request includes a product name, product attributes and customer information, wherein the product attributes include product specifications, product materials and product quantities, perform semantic analysis on the product name, extract product name keywords through natural language processing technology, expand the product name keywords through preset synonym expansion rules, obtain a product name keyword set, standardize the product attributes, map them through an attribute mapping matrix and determine product attribute features by calculating similarity, determine customer quotation preference features based on the customer information, combine with customer portrait data for fusion, and construct a customer comprehensive feature vector. The data obtained by comprehensive calculation obtains a multi-dimensional quotation request feature vector including: Receive a quotation request uploaded by a client, wherein the quotation request includes a product name, product attributes, and customer information, and the product attributes include product specifications, product materials, and product quantity; For the product name, the product name is segmented and tagged with parts of speech by a word segmentation tool, and the key entities in the product name are identified in combination with a pre-trained named entity recognition model to generate an entity recognition result, and the keywords in the product name are extracted in combination with the part-of-speech tagging result to obtain an initial keyword set, and a synonym word list is constructed based on the initial keyword set through a synonym expansion mechanism, and for each element in the initial keyword set, a match is performed in the synonym word list to determine the corresponding synonym, and the matched synonyms are added to the initial keyword set to obtain the product name keyword set; For each commodity type, a corresponding attribute mapping matrix is ​​constructed, wherein the rows of the attribute mapping matrix represent standardized attributes, the columns represent non-standardized attributes, and the elements in the attribute mapping matrix represent the mapping relationship between two commodity attributes. For the commodity attributes in the quotation request, the standardized attributes of each commodity attribute in the attribute mapping matrix are determined and the non-standardized attributes are converted into standardized attributes. The converted standardized attributes are converted into vector representations, and the commodity attribute features corresponding to the current commodity attributes are extracted through one-hot encoding; Based on the customer information, according to the customer's historical quotation request data, determine the customer quotation preference characteristics corresponding to the current customer through an association rule mining algorithm, obtain customer portrait data including basic information, consumption behavior and credit rating from the customer information system, and fuse the customer quotation preference characteristics with the customer portrait data to obtain the customer comprehensive feature vector; The product name keyword set, the product attribute features and the customer comprehensive feature vector are concatenated to obtain the multi-dimensional quotation request feature vector.

3. The method according to claim 2, characterized in that The customer quotation preference characteristics corresponding to the current customer are determined by the association rule mining algorithm as shown in the following formula: Where p represents the vector representation of customer quotation preference features, n represents the total number of rules obtained by the association rule mining algorithm, d represents a single quotation request in the customer historical quotation request dataset D, and X i represents the premise part of the i-th association rule, Y i represents the conclusion part of the i-th association rule, v i The binary vector representation of the preference attribute combination corresponding to the i-th rule.

4. The method according to claim 1, characterized in that: Based on the multi-dimensional quotation request feature vector, a combined retrieval condition based on semantic similarity and attribute similarity is constructed, and the target quotation data matching the combined retrieval condition is retrieved in combination with the improved inverted index technology, and the feature vector of the target quotation data is extracted, including supplier characteristics, quotation characteristics, supply cycle characteristics and scoring characteristics, and the quotation characteristics and the supply cycle characteristics are weighted by adaptive weights, and the scoring characteristics are attenuated in combination with a preset time attenuation function. The processed target quotation data is screened through a preset multi-criteria decision model, and a comprehensive matching degree matrix is ​​generated by calculation and screened to obtain high-quality target quotation data, including: Based on the multi-dimensional quotation request feature vector, original quotation data corresponding to the quotation request is defined, the quotation request and the original quotation data are preprocessed to generate a normalized text representation, words in the normalized text representation are converted into word vectors through a preset word embedding model, a vector representation of the normalized text is obtained through weighted averaging, and a cosine similarity between the quotation request and the original quotation data is calculated and used as a semantic similarity; Normalizing the structured attributes of the quotation request and the original quotation data and mapping them to the interval of 0 to 1, calculating the relative difference between the attribute value of the quotation request and the attribute value of the original quotation data for numerical attributes, determining whether the two attribute values ​​completely match for categorical attribute values, weighting and summing the similarity of each attribute to obtain a comprehensive attribute similarity, determining the comprehensive similarity through weighted summation based on the semantic similarity and the comprehensive attribute similarity, and using the comprehensive similarity as a measure of the combined search condition; Based on the combined retrieval condition, the candidate quotation data is determined by an improved inverted index technology, the keyword clusters and attribute clusters of the candidate quotation data are determined by a clustering algorithm, a cluster identification code is assigned to each keyword cluster and attribute cluster and a corresponding inverted index is established, the keyword vector and attribute vector of the candidate quotation data are determined by word embedding and attribute normalization processing, the similarity between each keyword vector and the keyword cluster center vector and the similarity between each attribute vector and the attribute cluster center vector are calculated respectively, the top ten keyword clusters and the top ten attribute clusters with the highest similarity are selected and the corresponding candidate quotation data sets are obtained, the combined retrieval condition is calculated and sorted for the selected candidate quotation data, and the candidate quotation data with the highest comprehensive similarity is used as the target quotation data; Feature extraction is performed on the target quotation data to obtain feature vectors corresponding to supplier features, quotation features, supply cycle features and scoring features, weighted processing is performed on the quotation features and the supply cycle features through an adaptive weight mechanism, the scoring features are decayed through a time decay function, the processed target quotation data are screened in combination with a pre-set multi-criteria decision model, the comprehensive matching degree matrix corresponding to each target quotation data is calculated, the evaluation score corresponding to each target quotation data is calculated based on the comprehensive matching degree matrix and compared with a preset evaluation score threshold, if the evaluation score is greater than the evaluation score threshold, the current target quotation data is retained as high-quality target quotation data, otherwise it is deleted, and multiple high-quality target quotation data are obtained.

5. The method according to claim 4, characterized in that Assign weights to each attribute as shown in the following formula: Among them, ω i represents the weight of the i-th attribute, λ1 represents the range weight, r i represents the value range of the i-th attribute, λ2 represents the stability weight, and s i represents the stability of the i-th attribute, λ3 represents the discrimination weight, and d i represents the distinguishability of the i-th attribute, r j Indicates the value range of the jth attribute, s j represents the stability of the jth attribute, d j represents the distinguishability of the jth attribute, and n represents the total number of attributes.

6. The method according to claim 1, characterized in that Sorting the high-quality target quotation data, determining the best quotation data according to the sorting result, pushing the best quotation data to the client based on the customer information, receiving the feedback information and feedback score of the best quotation data from the client, if the feedback information is to accept the best quotation data, generating corresponding order information and sending it to the supplier related to the best quotation data, if the feedback information is to reject the best quotation data, updating the best quotation data and re-pushing it includes: Based on the high-quality target quotation data, the data are sorted by an improved approximate ideal solution sorting method to generate a decision matrix for the high-quality target quotation data, the decision matrix is ​​standardized to generate a standard matrix, a weighted standardized decision matrix is ​​calculated in combination with customer weight preferences, a positive ideal solution and a negative ideal solution are determined based on the weighted standardized decision matrix, a weighted Euclidean distance between each high-quality target quotation data and the positive ideal solution and the negative ideal solution is calculated, and a relative closeness coefficient is defined based on the weighted Euclidean distance; Taking the positive ideal solution as a reference sequence and the high-quality target quotation data as a comparison sequence, calculating the correlation coefficient between the comparison sequence and the reference sequence, calculating the grey correlation degree based on the correlation coefficient, calculating the comprehensive evaluation value of each high-quality target quotation data based on the grey correlation degree and the relative closeness coefficient, sorting in descending order based on the comprehensive evaluation value, and taking the high-quality target quotation data ranked first as the optimal quotation data; Based on the customer information, the best quotation data is sent to the client and the feedback information and feedback score of the best quotation data from the client are received. If the feedback information is the best quotation data received, the corresponding order information is generated and sent to the corresponding supplier for confirmation and execution, the current feedback information is added to the customer comprehensive feature vector and the comprehensive evaluation value is updated based on the feedback score corresponding to the current feedback information; If the feedback information is to reject the optimal quotation data, a penalty item coefficient is defined and the penalty item of the comprehensive matching degree matrix is ​​adjusted and updated. The optimal quotation data is updated according to the adjusted comprehensive matching degree matrix and pushed to the client again.

7. The method according to claim 6, characterized in that The grey correlation degree is calculated based on the correlation coefficient as shown in the following formula: Among them, ψ m represents the grey correlation degree of the mth high-quality target quotation data, M represents the number of evaluation indicators, x 0n represents the value of the reference sequence on the nth evaluation index, x mn represents the value of the mth comparison series on the nth evaluation index, ρ represents the resolution coefficient, It represents the maximum absolute difference between all comparison series and reference series in each evaluation index. It represents the minimum absolute difference between all comparison series and reference series in each evaluation index.

8. A feedback push system based on background quotation data, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to receive a quotation request uploaded by a client, wherein the quotation request includes a product name, product attributes and customer information, the product attributes include product specifications, product materials and product quantities, perform semantic analysis on the product name, extract product name keywords through natural language processing technology, expand the product name keywords through preset synonym expansion rules to obtain a product name keyword set, perform standardization on the product attributes, map them through an attribute mapping matrix and determine product attribute characteristics by calculating similarity, determine customer quotation preference characteristics according to the customer information, combine with customer portrait data for fusion, construct a customer comprehensive feature vector, and obtain a multi-dimensional quotation request feature vector based on the data obtained through comprehensive calculation; The second unit is used to construct a combined search condition based on semantic similarity and attribute similarity based on the multi-dimensional quotation request feature vector, retrieve target quotation data matching the combined search condition in combination with the improved inverted index technology, extract the feature vector of the target quotation data, including supplier characteristics, quotation characteristics, delivery cycle characteristics and scoring characteristics, weight the quotation characteristics and the delivery cycle characteristics by adaptive weights, perform attenuation processing on the scoring characteristics in combination with a preset time attenuation function, screen the processed target quotation data by a preset multi-criteria decision model, generate a comprehensive matching degree matrix by calculation, and obtain high-quality target quotation data by screening; The third unit is used to sort the high-quality target quotation data, determine the best quotation data according to the sorting result, push the best quotation data to the client based on customer information, receive feedback information and feedback score from the client on the best quotation data, and if the feedback information is to accept the best quotation data, generate corresponding order information and send it to the supplier related to the best quotation data; if the feedback information is to reject the best quotation data, update the best quotation data and push it again.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images