User queue optimization method and system based on historical consultation data

Abstract

The present application relates to the technical field of message interface, in particular to a user queue optimization method and system based on historical consultation data, which comprises the following steps: querying background data, and using historical transaction data to establish a vector matrix, which is used to classify and arrange the background data; extracting keywords of current user chat data, fitting and matching multiple selected keywords with the vector matrix, if matching, outputting an actual fitting signal, if not matching, associating the keywords as twin words to re-match with the vector matrix, and calculating the fuzzy values of the twin words and the keywords; establishing a sorting rule to perform sorting, in view of the problem that the traditional user queue does not have a focus on user sorting and display, the system performs fitting retrieval on the keywords of user information and the background data, so as to re-sort the user list according to the fitting degree, and then facilitate the customer service personnel to focus more on communicating with users who are easy to transact, thereby improving the overall transaction rate.

Description

Technical Field

[0001] This invention relates to the field of messaging interface technology, and more specifically, to a method and system for optimizing user queues based on historical consultation data. Background Technology

[0002] In the process of consulting and customizing non-standard industrial parts, customer service personnel need to ask and verify with users multiple times to determine the specific type of non-standard parts. However, in actual work, customer service staff usually need to consult and connect with multiple users. However, the latest message pop-up mechanism used by traditional communication software makes the order of multiple users in the user list unrelated to the backend data, and it is impossible to intuitively show the current user's probability of closing a deal from the user's sorting position. Summary of the Invention

[0003] The purpose of this invention is to provide a user queue optimization method and system based on historical consultation data to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, this invention provides a user queue optimization method based on historical consultation data, comprising the following steps:

[0005] Query backend data and use historical transaction data to build a vector matrix, which is used to compare with the keywords selected by the user;

[0006] Extract keywords from the current user's chat data, fit and match multiple selected keywords with a vector matrix. If a match is found, output the actual fitted signal. If no match is found, associate the keywords with twins and rematch them with the vector matrix, output the fuzzy fitted signal, and calculate the fuzzy value of the twins and keywords.

[0007] Establish sorting rules and perform sorting, including:

[0008] The actual fitted signal has a higher priority than the fuzzy fitted signal.

[0009] When only the actual fitted signal is output, multiple users are displayed at the top of the user list;

[0010] When only the fuzzy fitted signal is output, the fuzzy value is calculated, and multiple users are sorted from high to low according to the fuzzy value.

[0011] As a further improvement to this technical solution, the establishment of the vector matrix includes performing tree-like classification on the background data, extracting multiple standard keywords, and using the multiple standard keywords as the coordinate axes of the vector matrix;

[0012] Wherein, the standard keywords of the same category serve as the coordinate axes of the same dimension in the vector matrix, the coordinate axes corresponding to the standard keywords of different categories intersect perpendicularly, and the intersection point is the background data that simultaneously meets multiple corresponding standard keywords.

[0013] As a further improvement of this technical solution, the extracting the keywords of the current user chat data includes the following steps:

[0014] Character recognition is performed on the user chat data to generate text information of the user chat data, common meaningless words such as "de" and "shi" in the text information are deleted, the text information phrase is split into a plurality of independent phrases by taking the deletion positions as separation points, characters in the independent phrases are collected and counted, and words that repeatedly appear in the independent phrases are recorded and marked as selected keywords.

[0015] As a further improvement of this technical solution, the step of taking a plurality of standard keywords as coordinate axes of the vector matrix further includes the following steps:

[0016] A standard keyword temporary storage table for duplicate checking is established, and a plurality of standard keywords are stored into the standard keyword temporary storage table one by one. Wherein, when storing a standard keyword into the standard keyword temporary storage table, the already stored data in the standard keyword temporary storage table is retrieved; if the same standard keyword already exists in the standard keyword temporary storage table, the standard keyword is skipped; if the standard keyword is not stored in the standard keyword temporary storage table, the keyword is stored into the standard keyword temporary storage table, and then the vector matrix is established using the standard keywords in the standard keyword temporary storage table.

[0017] As a further improvement of this technical solution, the fitting and matching of the plurality of selected keywords with the vector matrix includes the following steps:

[0018] The plurality of selected keywords are compared with the standard keyword combinations at the intersection points of the coordinate axes in the vector matrix; if the selected keywords are consistent with the standard keyword combinations at the intersection points of the coordinate axes in the vector matrix, an actual fitting signal is output; if not, association is performed on the selected keywords to generate twin words.

[0019] As a further improvement of this technical solution, the associating the selected keywords to generate twin words includes the following steps:

[0020] The Sentence-BERT algorithm is used to perform association calculation on the selected keywords and generate twin words corresponding to the selected keywords, the twin words are input into the vector matrix one by one and compared with the standard keyword combinations at the intersection points of the coordinate axes in the vector matrix; if the twin words are consistent with the standard keyword combinations at the intersection points, a fuzzy fitting signal is output, and a fuzzy value is calculated according to the successfully matched twin words and the selected keywords.

[0021] As a further improvement to this technical solution, the step of calculating the fuzzy value based on the successfully compared twins and selected keywords includes the following steps:

[0022] The selected keywords and twin words are input into the SBERT model. The SBERT model generates a fixed-length numerical vector for the selected keywords and twin words. The directions of the two vectors are compared to calculate their cosine similarity. The absolute value of the cosine similarity is taken and used as the fuzzy value.

[0023] As a further improvement to this technical solution, establishing sorting rules and performing sorting includes the following steps:

[0024] If multiple users only output the actual fitted signal, sort them according to the last time the information was sent by the users, and display the user whose last information sending time is closest to the current time at the top of the user list;

[0025] If the user list contains both users who output actual fitted signals and users who output simulated fitted signals, sort the users who output actual fitted signals at the top of the user list according to the last time the information was sent, sort the users who output simulated fitted signals in descending order according to the fuzziness value, and place the users who output simulated fitted signals below the users who output actual fitted signals.

[0026] A user queue optimization system based on historical consultation data, applied to the user queue optimization method based on historical consultation data, includes a system page, a tree-structured storage, a vector matrix matching module, and a sorting execution module;

[0027] The system page includes a user list display box, a keyword selection box, a feedback result display box, and a fitting button. The user list display box is used to display the optimized and sorted user list. The keyword selection box is used to display the selected keywords. The feedback result display box is used to display the actual fitted signal or fuzzy value result corresponding to a user in the user list after selection. The fitting button is used to control the progress of comparing the selected keywords with the vector matrix.

[0028] The tree-structured storage is used to store historical background data and a vector matrix based on historical transaction probabilities;

[0029] The vector matrix matching module is used to extract keywords from multiple user chat data in the user list, and to build a keyword selection box for selecting keywords. It fits multiple selected keywords with the vector matrix. If they match, the actual fitted signal is output. If they do not match, the keywords are associated with twins and re-matched with the vector matrix. The fuzzy fitted signal is output, and the fuzzy value of the twins and keywords is calculated.

[0030] The sorting execution module is used to sort the users in the user list based on the actual fitted signal or fuzzy value calculated by the user comparison, and follows the rule that the actual fitted signal has higher sorting priority than the fuzzy fitted signal.

[0031] As a further improvement to this technical solution, the optimization system also includes a front-end display module, which is used to display system pages and has a communication association with the tree-structured storage.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] 1. In this user queue optimization method and system based on historical consultation data, the structured and standardized processing of user consultation information is achieved by querying historical transaction data in the background and constructing a vector matrix. Then, by extracting keywords from the current user chat data and fitting and matching them with the vector matrix (including direct matching and semantic association matching based on the Sentence-BERT algorithm), users who match the historical high transaction probability model can be intelligently identified. Based on this, the priority ranking rule (the actual fitted signal takes precedence over the fuzzy fitted signal, and is optimized in combination with the last sending time) makes the user list no longer mechanically arranged in chronological order, but dynamically rearranged according to the potential transaction probability.

[0034] 2. In this user queue optimization method and system based on historical consultation data, efficient classification and matching of user consultation information is achieved by querying backend data and constructing a vector matrix using historical transaction records. By extracting keywords from user chat data and fitting and matching them with the vector matrix, high-intent users can be accurately identified. When a precise match cannot be achieved, twin words are generated through semantic association and fuzzy matching degree is calculated, which significantly improves the accuracy and flexibility of user intent recognition. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the process of the present invention.

[0036] Figure 2 This is a schematic diagram of the vector matrix in this invention;

[0037] Figure 3 This is a schematic diagram of the optimized system page in this invention. Detailed Implementation

[0038] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Example 1

[0040] Please see Figures 1-3 As shown, this embodiment provides a user queue optimization method and system based on historical consultation data. The present invention provides a user queue optimization method based on historical consultation data, including the following steps:

[0041] S1. Query the backend data and use historical transaction data to build a vector matrix. The vector matrix is ​​used to compare with the keywords selected by the user. In the process of building the vector matrix, the backend data needs to be classified into a tree structure. Multiple standard keywords are extracted from the backend data and used as the coordinate axes of the vector matrix.

[0042] When performing tree-like classification on the backend data, the tree-like classification process needs to be described in detail. In the backend data, a single backend data unit includes historical successful transaction records and the standard keywords corresponding to the production records. The backend data is classified according to the standard keywords of the production records (backend data units with the same standard keywords are divided into the same category). After the standard keyword classification, the backend data unit is divided into several categories, and the category name is named with the same standard keyword in the category.

[0043] After a category is created, a clustering algorithm is used to calculate and generate upper-level categories, and the upper-level categories are named until the number of upper-level categories is one.

[0044] It should be noted that clustering algorithms include the K-Means algorithm. The K-Means algorithm calculates the upper-level classification data of the current classification data. The specific steps are as follows:

[0045] Randomly select several category names as initial cluster centers; calculate the distance from each category name to each centroid (usually using Euclidean distance), and assign it to the cluster containing the nearest centroid. This step achieves upward summarization and generalization of sample category names by assigning the calculated centroids, which are common features of several category names, to the cluster containing the nearest centroid.

[0046] The average value of all sample points in each cluster is recalculated and used as the new centroid. This step corrects the calculated centroid to more accurately express the common features of several classification names.

[0047] Repeat the above steps until the position of the centroid no longer changes significantly. When the centroid no longer changes, it means that the centroid has stabilized and the upper-level classification summary name has converged. At this time, use the centroid content as the upper-level classification name and include several classification names as samples in the upper-level classification. Then repeat the above steps again to summarize and generalize layer by layer until the number of the top-level classifications is one and stop the calculation.

[0048] The above calculation method can be used to classify and store the backend data in a tree structure. When generating the vector matrix, standard keywords of the same type can be used as different coordinate axes of the same dimension according to the classification combination.

[0049] Standard keywords from different categories are read to form the data basis for building the vector matrix. However, there may be duplicate content among the standard keywords in different categories (some standard keywords may satisfy two classification criteria at the same time, resulting in the same standard keywords in different categories). In order to make the vector matrix more concise and improve the efficiency of the subsequent comparison process, a temporary standard keyword table is established for deduplication when refining standard keywords. Multiple standard keywords are stored into the temporary standard keyword table one by one. When storing standard keywords into the temporary standard keyword table, the data already stored in the temporary standard keyword table is retrieved.

[0050] If the same standard keyword already exists in the standard keyword temporary storage table, skip that standard keyword; if the standard keyword does not exist in the standard keyword temporary storage table, add that keyword to the standard keyword temporary storage table. Then, construct a vector matrix using the standard keywords in the standard keyword temporary storage table. By carefully filtering the standard keyword temporary storage table, duplicate keywords can be removed, thereby reducing the total number of coordinate axes in the vector matrix. The coordinates of the same standard keyword will not affect the comparison process, but will only reduce the efficiency of comparing the selected keywords with the vector matrix. Therefore, by removing duplicate standard keywords, the effect of simplifying the vector matrix can be achieved.

[0051] After deduplication of standard keywords, such as Figure 2 As shown, the standard keywords of the same category are used as the coordinate axes of the same dimension in the vector matrix. The coordinate axes corresponding to the standard keywords of different categories intersect each other perpendicularly, and the intersection point is the backend data that simultaneously meets multiple corresponding standard keywords.

[0052] like Figure 2as shown in the drawing, the vector matrix is a two-dimensional vector matrix formed by mutually combining standard keywords of two classifications. In actual comparison, vector matrices of different dimensions can be established according to the type and scale of back-end data (a three-dimensional vector matrix can be established using standard keywords of three classifications). However, it should be noted that increasing the dimensions will cause the number of axis intersection points in the vector matrix to increase exponentially, so the maximum dimension of the vector matrix is determined by the operating performance of the vector matrix matching module.

[0053] S2. Extract keywords of current user chat data: generate text information of the user chat data through graphic recognition on the user chat data, delete common meaningless words such as "de" and "shi" in the text information, so that the text information sentence can be decomposed into a plurality of independent phrases with the deletion positions as separation points, perform summary statistics on characters in the independent phrases, and record repeatedly appearing words in the independent phrases and mark them as selected keywords;

[0054] It should be supplemented that the step of generating the text information of the user chat data through graphic recognition on the user chat data may adopt a manner of performing OCR recognition on the screen of the communication software page to extract and capture the text information content;

[0055] After obtaining the selected keywords in the user text information, perform fitting matching between the plurality of selected keywords and the vector matrix. If the matching is successful, an actual fitting signal is output; if the matching is unsuccessful, the keywords are associated as twin words to re-match with the vector matrix, a fuzzy fitting signal is output, and the fuzzy value of the twin words and the keywords is calculated. In the matching process, the plurality of selected keywords are compared with the standard keyword combinations at the axis intersection points in the vector matrix. If the selected keywords are consistent with the standard keyword combinations at the axis intersection points in the vector matrix, the actual fitting signal is output; if not, the step of associating the selected keywords to generate twin words is executed;

[0056] Herein, the method for associating the selected keywords to generate twin words is to use the Sentence-BERT algorithm to perform association calculation on the selected keywords, generate twin words corresponding to the selected keywords, input the twin words into the vector matrix one by one and compare them with the standard keyword combinations at the axis intersection points in the vector matrix. If the twin words are consistent with the standard keyword combinations at the intersection points, the fuzzy fitting signal is output, and the fuzzy value (the fuzzy value is of a floating-point type and ranges from zero to one) is calculated according to the successfully compared twin words and the selected keywords;

[0057] Here is a brief supplement to the Sentence-BERT algorithm. When the selected keyword is input into the Sentence-BERT algorithm, the selected keyword will be stored in the BERT encoder. The BERT encoder will generate several context-related hidden state vectors for each selected keyword, calculate the average value of all hidden state vectors, and take the maximum value of all hidden state vectors in each dimension to obtain the word group corresponding to the maximum value of the hidden state vector in each dimension. This word group is the twin word of the selected keyword.

[0058] Specifically, the method for calculating the fuzzy value is as follows: input the selected keyword and the twin word into the SBERT model respectively. The SBERT model generates a fixed-length numerical vector for the selected keyword and the twin word. Compare the directions of the two vectors to calculate their cosine similarity (cosine similarity is a floating-point type, ranging from negative one to one). Take the absolute value of the cosine similarity and use the absolute value of the cosine similarity as the fuzzy value.

[0059] The above matching process can generate actual fitted signal results or fuzzy fitted signal results and fuzzy value results. Based on the numerical information output after calculation, the degree of matching between the user's selected keywords and the background data can be judged.

[0060] To prevent extreme keyword selection from causing the backend data to enter an infinite matching loop, a minimum fuzzy value needs to be limited. When the user's selected keyword differs too much from the backend data, the twin words generated after several association algorithms still cannot match the coordinate axis intersections in the backend data. Therefore, after a twin word match fails, the fuzzy value of the current twin word is calculated. When the fuzzy value is lower than a certain value, the association calculation stops and the result of no match is output. By setting a minimum fuzzy value, the association algorithm can be prevented from entering an infinite loop.

[0061] After obtaining the actual fitted signal result or the fuzzy fitted signal result and fuzzy value result:

[0062] S3. Establish sorting rules and perform sorting, including:

[0063] The actual fitted signal has a higher priority than the fuzzy fitted signal.

[0064] When only the actual fitted signal is output, multiple users are displayed at the top of the user list;

[0065] When only the fuzzy fitted signal is output, the fuzzy value is calculated, and multiple users are sorted from high to low according to the fuzzy value;

[0066] Furthermore, if multiple users only output the actual fitted signal, sort them according to the last time the information was sent by the multiple users, and display the user whose last information sending time is closest to the current time at the top of the user list;

[0067] If the user list contains both users who output actual fitted signals and users who output simulated fitted signals, sort the users who output actual fitted signals at the top of the user list according to the last time the information was sent, sort the users who output simulated fitted signals in descending order according to the fuzziness value, and place the users who output simulated fitted signals below the users who output actual fitted signals.

[0068] Based on the above sorting rules, the user list can be reordered on the system page, and the users with the highest matching degree can be displayed at the top of the user list. This allows customer service to intuitively judge the probability of a user making a purchase and the degree of matching between the user's needs and existing production records based on the user's sorting order, thus optimizing the user queue.

[0069] Example 2

[0070] Please see Figures 1-3 As shown, this embodiment provides a user queue optimization system based on historical consultation data, which is applied to the user queue optimization method based on historical consultation data in Embodiment 1. It includes a system page, a tree-structured memory, a vector matrix matching module, a sorting execution module, and a front-end display module. The front-end display module is used to display the system page, and the front-end display module has a communication association with the tree-structured memory.

[0071] The system page includes a user list display box, a keyword selection box, a feedback result display box, and a fitting button. The user list display box is used to display the optimized and sorted user list. The keyword selection box is used to display the selected keywords. The feedback result display box is used to display the actual fitted signal or fuzzy value result corresponding to a user selected from the user list. The fitting button is used to trigger the front-end display module to send information to the tree-structured memory.

[0072] In practical use, the front-end data module and the tree-structured storage typically use the Internet for data transmission. There is a risk that information may be distorted during transmission, which could lead to invalid calculations by the vector matrix matching module.

[0073] Therefore, when the front-end display module sends the keyword selection information, a verification code based on the selected keyword is added to the end of the information. After receiving the information, the tree-structured storage verifies the information content according to the verification code. If the verification passes, it is sent to the vector matrix matching module for comparison and calculation. If the verification fails, the tree-structured storage sends a repeat sending command to the front-end data module.

[0074] The verification code uses a cyclic redundancy check method. The selected keyword is used as sample data and converted into binary data. Then, modulo 2 division is performed on it, and the remainder is the verification code. The verification code corresponds one-to-one with the selected keyword and can be used to judge the integrity of the selected verification code information. When verifying the selected keyword, the above method is used to calculate the verification code of the selected keyword and compare it with the verification code carried in the sent information. If they are the same, it means that the information is complete.

[0075] Specifically, modulo-2 division is a binary division method based on XOR operation that does not consider carry or borrow. Starting from the most significant bit of the dividend, it takes the same number of bits as the divisor and only considers the most significant bit of the current part of the dividend. If it is 1, the quotient is 1; if it is 0, the quotient is 0. After the calculation is completed, all the quotient results are combined to form the verification code.

[0076] The tree-structured storage is used to store historical background data and a vector matrix based on historical transaction probabilities;

[0077] The vector matrix matching module is used to extract keywords from multiple user chat data in the user list, and to build a keyword selection box for selecting keywords. It fits multiple selected keywords with the vector matrix. If they match, the actual fitted signal is output. If they do not match, the keywords are associated with twins and re-matched with the vector matrix. The fuzzy fitted signal is output, and the fuzzy value of the twins and keywords is calculated.

[0078] The sorting execution module is used to sort the users in the user list based on the actual fitted signal or fuzzy value calculated by the user comparison, and follows the rule that the actual fitted signal has higher sorting priority than the fuzzy fitted signal.

[0079] To prevent keyword leaks during information transmission over the internet, encryption modules can be added to the front-end display module and the tree-structured storage. Before the front-end display module sends information, it encrypts the information to form ciphertext, which is then sent to the tree-structured storage. The tree-structured storage then decrypts the ciphertext, thus achieving the effect of encrypted information transmission.

[0080] Furthermore, the method for generating ciphertext is AES encryption. AES encryption is a symmetric encryption method that can restore ciphertext to plaintext data based on ciphertext and key. By pre-storing the same key data in the front-end display module and the tree-structured storage, the effect of encrypting and decrypting data conversion can be achieved.

[0081] Based on the above embodiments, the principle of use of this application is as follows:

[0082] Query background data, perform tree classification on historical transaction data, and extract multiple standard keywords (such as product type, processing requirements, etc.). During classification, a clustering algorithm (such as K-Means) is used for layer-by-layer induction to form a tree structure.

[0083] Establish a vector matrix: take standard keywords as coordinate axes, keywords of the same category as coordinate axes of the same dimension, and keywords of different categories correspond to vertically intersecting coordinate axes, where the intersection point represents historical data that simultaneously meets multiple standard keywords. To reduce redundancy, the matrix is constructed after deduplication through a temporary storage table.

[0084] The vector matrix is used as a matching reference, and its dimensions can be adjusted according to the data scale (such as two-dimensional or three-dimensional).

[0085] Convert the user's chat content into text information through graphic recognition (such as OCR).

[0086] Preprocess the text: delete meaningless words (such as "de", "shi"), and decompose the sentence into independent phrases.

[0087] Count the repeatedly appearing words in the phrases, mark them as "selected keywords" as the input for matching.

[0088] There are two results in the matching process as follows:

[0089] Compare the selected keywords with the standard keyword combinations at the intersections of the vector matrix. If they are completely consistent, output an "actual fitting signal" (indicating a high matching degree);

[0090] If there is no match, use the Sentence-BERT algorithm to associate the selected keywords to generate semantically similar "twin words", then match the twin words with the vector matrix. If they are consistent, output a "fuzzy fitting signal" and calculate a fuzzy value; wherein the fuzzy value is calculated by converting the selected keywords and the twin words into numerical vectors through the SBERT model, and taking the absolute value of the cosine similarity as the fuzzy value (between 0 and 1), where a higher value indicates closer semantics.

[0091] After obtaining the output results, sort the user list according to the sorting rules. In the sorting process, follow the rule that the actual fitting signal has a higher sorting priority than the fuzzy fitting signal; when only the actual fitting signal is output, multiple users are pinned to the top of the user list for display; when only the fuzzy fitting signal is output, the calculated fuzzy value is obtained, and multiple users are sorted in descending order of the fuzzy value.

[0092] This invention provides an intelligent optimization scheme for user queues based on historical consultation data. Its core value lies in transforming the traditional time-sequential user queue into a dynamic sorting system based on the strength of the transaction intention through data-driven and semantic analysis technologies, thereby significantly improving customer service efficiency and overall conversion rate.

[0093] The system then extracts keywords from user chat content in real time. Instead of simply matching them with historical data, it innovatively employs a semantic association matching mechanism based on Sentence-BERT. When a literal match fails, the system intelligently generates semantically similar "twin words" for rematching and calculates a quantifiable "fuzzy value" to accurately assess the degree of matching. This greatly enhances the system's ability to understand diverse and colloquial expressions, solving the pain point of rigid matching in traditional systems.

[0094] The system reorders the user queue based on a clear priority rule: users who obtain an "actual fitted signal" (perfect match) always have priority over users who obtain an "ambiguous fitted signal" (semantic match); within the same type of signal, the order is further optimized based on "last sending time" and "ambiguity value". This allows customer service personnel to identify the highest-value users at a glance from the system interface.

[0095] The significant advantage of this invention lies in the fact that it is not merely a single technological improvement, but a systematic solution integrating intelligent data processing, natural language understanding, and business goal optimization. It successfully transforms dormant historical data into valuable decision-making assets, liberating customer service staff from passive responses and tedious searches, allowing them to proactively focus on efficient transactions, ultimately enabling the customer service system to leap from a "tool" to an "intelligent assistant."

[0096] In terms of technical implementation, the system first performs in-depth mining and tree-like classification of historical transaction data in the background to construct a structured vector matrix with standard keywords as coordinate axes. This matrix serves as a precise measure of user intent.

[0097] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A user queue optimization method based on historical consultation data, characterized in that, Includes the following steps: The system queries the backend data. Each backend data unit includes historical successful transaction records and the standard keywords corresponding to those records. A vector matrix is ​​then built using the historical transaction data. This vector matrix is ​​used to compare the data with the keywords selected by the user. Extract keywords from the current user's chat data, fit and match multiple selected keywords with a vector matrix. If a match is found, output the actual fitted signal. If no match is found, associate the keywords with twins and rematch them with the vector matrix, output the fuzzy fitted signal, and calculate the fuzzy value of the twins and keywords. Establish sorting rules and perform sorting, including: The actual fitted signal has a higher priority than the fuzzy fitted signal. When only the actual fitted signal is output, multiple users are displayed at the top of the user list; When only the fuzzy fitted signal is output, the fuzzy value is calculated, and multiple users are sorted from high to low according to the fuzzy value.

2. The user queue optimization method based on historical consultation data according to claim 1, characterized in that: The establishment of the vector matrix includes performing tree-like classification on the background data, extracting multiple standard keywords, and using the multiple standard keywords as the coordinate axes of the vector matrix. In this system, standard keywords of the same category are used as coordinate axes of the same dimension in the vector matrix. The coordinate axes corresponding to standard keywords of different categories intersect each other perpendicularly, and the intersection point is the backend data that simultaneously meets multiple corresponding standard keywords.

3. The user queue optimization method based on historical consultation data according to claim 1, characterized in that: The process of extracting keywords from the current user's chat data includes the following steps: By generating text information from user chat data through graphical recognition, meaningless words in the text information are deleted. The text information phrases are decomposed into multiple independent phrases using the deletion position as the delimiter. The characters in the independent phrases are summarized and statistically analyzed. Repeated words in the independent phrases are recorded and marked as selected keywords.

4. The user queue optimization method based on historical consultation data according to claim 2, characterized in that: The method of using multiple standard keywords as coordinate axes of a vector matrix also includes the following steps: Establish a standard keyword temporary table for deduplication. Store multiple standard keywords into the standard keyword temporary table one by one. When storing standard keywords into the standard keyword temporary table, search the data already stored in the standard keyword temporary table. If the same standard keyword is already stored in the standard keyword temporary table, skip that standard keyword. If the standard keyword is not stored in the standard keyword temporary table, store that keyword into the standard keyword temporary table. Then, construct a vector matrix using the standard keywords in the standard keyword temporary table.

5. The user queue optimization method based on historical consultation data according to claim 1, characterized in that: The process of matching multiple selected keywords with a vector matrix includes the following steps: The selected keywords are compared with the standard keyword combinations at the intersection points of the coordinate axes in the vector matrix. If the selected keywords match the standard keyword combinations at the intersection points of the coordinate axes in the vector matrix, the actual fitted signal is output. If they do not match, twin words are generated by associating the selected keywords.

6. The user queue optimization method based on historical consultation data according to claim 5, characterized in that: The process of generating twin words from selected keywords includes the following steps: The Sentence-BERT algorithm is used to perform association calculations on the selected keywords and generate twin words corresponding to the selected keywords. The twin words are input into the vector matrix one by one and compared with the standard keyword combination at the intersection of the coordinate axes in the vector matrix. If the twin word matches the standard keyword combination at the intersection, the fuzzy fitting signal is output, and the fuzzy value is calculated based on the successfully matched twin word and the selected keyword.

7. The user queue optimization method based on historical consultation data according to claim 6, characterized in that: The calculation of the fuzzy value based on the successfully matched twins and selected keywords includes the following steps: The selected keywords and twin words are input into the SBERT model. The SBERT model generates a fixed-length numerical vector for the selected keywords and twin words. The directions of the two vectors are compared to calculate their cosine similarity. The absolute value of the cosine similarity is taken and used as the fuzzy value.

8. The user queue optimization method based on historical consultation data according to claim 1, characterized in that: The process of establishing sorting rules and performing sorting includes the following steps: If multiple users only output the actual fitted signal, sort them according to the last time the information was sent by the users, and display the user whose last information sending time is closest to the current time at the top of the user list; If the user list contains both users who output actual fitted signals and users who output fuzzy fitted signals, sort the users who output actual fitted signals at the top of the user list according to the last time the information was sent, sort the users who output fuzzy fitted signals in descending order according to the fuzziness value, and place the users who output fuzzy fitted signals below the users who output actual fitted signals.

9. A user queue optimization system based on historical consultation data, used to implement the user queue optimization method based on historical consultation data as described in any one of claims 1 to 8, characterized in that: This includes system pages, a tree-structured storage, a vector-matrix matching module, and a sorting execution module; The system page includes a user list display box, a keyword selection box, a feedback result display box, and a fitting button. The user list display box is used to display the optimized and sorted user list. The keyword selection box is used to display the selected keywords. The feedback result display box is used to display the actual fitted signal or fuzzy value result corresponding to a user in the user list after selection. The fitting button is used to control the progress of comparing the selected keywords with the vector matrix. The tree-structured storage is used to store historical background data and a vector matrix based on historical transaction probabilities; The vector matrix matching module is used to extract keywords from multiple user chat data in the user list, and to establish a keyword selection box for selecting keywords. It fits multiple selected keywords with the vector matrix. If they match, the actual fitted signal is output. If they do not match, the keywords are associated with twins and re-matched with the vector matrix. The fuzzy fitted signal is output, and the fuzzy value of the twins and keywords is calculated. The sorting execution module is used to sort the users in the user list according to the actual fitting signal or fuzzy value calculated by the user comparison, and follows the rule that the actual fitting signal has a higher sorting priority than the fuzzy fitting signal.

10. The user queue optimization system based on historical consultation data according to claim 9, characterized in that: The optimization system also includes a front-end display module, which is used to display system pages and has a communication association with the tree-structured storage.

Images