Network technology service consulting intelligent matching system based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based intelligent matching system for network technology service consultation, which relates to the field of artificial intelligence technology. The present invention first collects historical network technology problems and corresponding solutions through a data collection module, and extracts main and secondary keywords. The search heat calculation module calculates the search heat of keyword combinations, introduces parameters such as the number of searches, user feedback scores, and a time decay factor to construct a formula, and the database storage module sorts problems and solutions by heat and then stores them. When a user inputs a question, the user question matching module extracts keywords, determines combinations with higher search heat based on the main keywords, uses them as candidate data, converts the keywords into digital vectors to calculate similarity, and displays solutions with similarity higher than a threshold to the user according to the size of the similarity. The system can efficiently and accurately provide users with solutions to network technology problems, thereby improving the quality and efficiency of network technology service consultation.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based network technology service consultation intelligent matching system. Background Art

[0002] In today's digital age, network technology is widely used in various fields, and users encounter a variety of problems while using it. Traditional network technology service consultation methods often rely on manual customer service, which is inefficient and difficult to meet the needs of a large number of users. With the development of artificial intelligence technology, although some intelligent customer service has been applied to network technology service consultation, there are still deficiencies in the accuracy of problem matching and solution provision. Existing systems cannot fully utilize historical data and search popularity information, resulting in a poor match between the solutions users receive and their actual problems, and unable to effectively solve users' network technology problems. Therefore, a new intelligent matching system is needed to improve service quality and efficiency. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the present invention provides an artificial intelligence-based intelligent matching system for network technology service consultation, which solves the problem that the solutions obtained by users in actual network technology service consultation do not match the actual situation.

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: an artificial intelligence-based network technology service consultation intelligent matching system, comprising:

[0005] Search popularity calculation module, which calculates search popularity based on the search frequency of the main keyword and secondary keyword combination, user feedback score, search depth adjustment coefficient, and semantic similarity matrix;

[0006] A database storage module, which sorts network technology problems and their corresponding solutions according to search popularity and stores them in a search popularity database. The database structure includes a primary keyword plus a secondary keyword field and a corresponding solution field;

[0007] User question matching module, which extracts the main keywords and secondary keywords of the user input question, counts the total number of characters Total of the two, locates the combination about the main keyword in the search popularity database according to the main keyword, extracts the combination with a search popularity greater than Sth, calculates the total number of characters Totali of the main keyword and secondary keyword of this combination. If Totali <= Total, add this combination to the alternative comparison combinations, and obtain the corresponding digital vector according to the Unicode table for the combination of the main keyword and secondary keyword. If Totali < Total, add 0 after the obtained digital vector, and the number of added 0s is Total - Totali. The transformed digital vector is Vcompi. Also obtain the corresponding digital vector Vuser for the combination of the user's main keyword and secondary keyword according to the Unicode table, calculate the matching degree Gi between the vector Vcompi and Vuser. If Gi > Gth, add the solution to the network technology problem corresponding in the search popularity database to the display database, and take out the solution from the display database according to the size of the matching degree Gi to display to the user, where Gth is the set matching degree threshold, i ∈ [1, e], and e is the number of combinations with a search popularity greater than Sth.

[0008] As a further solution of the present invention, before the search popularity calculation module, there are also a data collection module and a main and secondary keyword extraction module. The data collection module is used to collect historical data, and the historical data includes network technology problems and corresponding solutions; the main and secondary keyword extraction module is used to perform word segmentation processing on the network technology problems, construct a co-occurrence graph between the word segments, and extract the main keyword and secondary keyword from the co-occurrence graph. The co-occurrence graph can reveal the relationship and weight between the keywords.

[0009] As a further solution of the present invention, the specific steps for extracting the main keyword and secondary keyword are as follows:

[0010] Use a word segmentation tool to split the problem description text into individual words or phrases;

[0011] ​​​​​​​​​​

[0015] Starting from the node corresponding to the main keyword, traverse the co-occurrence graph in hierarchical order. First, visit the direct neighbor nodes of the main keyword and record their weights. Then visit the neighbor nodes of these neighbor nodes, and so on. During the traversal process, for each newly visited node, calculate the product of the weights of all edges on the path from the main keyword to the node as the comprehensive weight of the node.

[0016] The attenuation factor ɑ is introduced to adjust the comprehensive weight, and the comprehensive weight Wi of all nodes directly and indirectly connected to the main keyword is obtained, i∈[1,n], n represents the number of all other nodes except the main keyword node;

[0017] Set a threshold Wth. If Wi>Wth, the word corresponding to the i-th node is used as the secondary keyword.

[0018] As a further solution of the present invention, after the primary keyword and the secondary keyword are combined to obtain the code point corresponding to each character according to the Unicode table, the code point needs to be converted into a decimal number.

[0019] As a further solution of the present invention, the search popularity of the main keyword and a single secondary keyword is calculated using the following formula, which is:

[0020]

[0021] Where F is the search frequency, R is the user feedback score, ranging from [-100, 100], and d is the search depth adjustment coefficient, ranging from [0, 1].

[0022] As a further solution of the present invention, the specific steps of calculating the search popularity of the main keyword and multiple secondary keywords are:

[0023] The word vector model is used to calculate the semantic similarity between every two sub-keywords. Assuming there are m sub-keywords K1, K2, ..., Km, the calculated semantic similarity matrix M is an m×m matrix, where Mij represents the semantic similarity between sub-keywords Ki and Kj.

[0024] Calculate the average value of all off-diagonal elements in the matrix, i≠j, the specific formula is: S total The value range of is [0,1];

[0025] According to the formula S'=S×(1+β×(S total -γ)) obtains the search popularity of the combination of the main keyword and multiple sub-keywords, where β is the weight adjustment coefficient and γ is the baseline value of the semantic association degree.

[0026] As a further solution of the present invention, when data sparsity occurs, the specific method of adjusting the search frequency by using the semantic similarity between secondary keywords is as follows:

[0027] Collect combinations Ct with F < λ, where Ct contains a main keyword and multiple sub-keywords, and filter out combinations Cr that are identical or similar to the main keyword of Ct, where r∈[1,p], p is the total number of combinations, and λ∈[0,1];

[0028] Calculate the semantic similarity Sr between Ct and Cr based on the word vector model, and find the combination Cr corresponding to max(Sr);

[0029] The search frequency Fr of Cr is calculated to obtain the estimated search frequency Festimate of Ct = max(Sr)×Fr.

[0030] The present invention provides an artificial intelligence-based intelligent matching system for network technology service consultation. Compared with the existing technology, it has the following advantages:

[0031] (1) The present invention constructs a co-occurrence graph through multiple network technology problems. The co-occurrence graph can clearly show the relationship between words. It is not just a simple co-occurrence problem, but also reflects the degree of semantic association. In addition, observing the structure of the co-occurrence graph can quickly understand the core content of the text, which helps to grasp and understand the overall content of the text;

[0032] (2) The present invention constructs a search popularity database based on the combination of primary keywords and secondary keywords, which can quickly locate the approximate scope of the user's input questions, reduce redundant calculations, and achieve efficient resource allocation. It further accurately presents solutions to users by calculating the matching degree, allowing users to see the most relevant information first, thereby improving user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a block diagram of the system principle of the present invention. DETAILED DESCRIPTION

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0035] like Figure 1 The present invention provides an artificial intelligence-based network technology service consultation intelligent matching system, comprising:

[0036] Data collection module: The core task of this module is to comprehensively and accurately collect historical network technology problems and their corresponding solutions, and extract keywords from these problems through natural language processing technology to provide data support for subsequent search popularity calculation and intelligent matching;

[0037] The collection channels are mainly divided into three aspects: open source data capture, user feedback records, and data sharing with cooperative enterprises and institutions;

[0038] For open source data capture, we use data scraping tools to capture relevant data from various public online technology forums, Q&A communities, technical blogs, and official vendor support forums. For example, on GitHub, we can capture programming-related questions and solutions, such as difficulties in using different programming languages, framework building issues, project deployment, and code optimization. The tool can automatically access these platforms regularly to extract questions and answers related to online technology.

[0039] Regarding user feedback records, the system automatically collects questions and feedback submitted by users during use. When users encounter network technical problems, they can submit problem descriptions through the feedback channels within the system, such as online forms, customer service chat windows, etc. At the same time, the system can also record the actual subsequent resolution of the user's questions. For example, a user of a certain network service provider encounters a slow network speed problem while using the network service. He reports the problem through the customer service chat window, and the customer service staff records the problem and provides a solution;

[0040] For data sharing with partner companies and institutions, establish cooperative relationships with other relevant companies and institutions to obtain the problems and solutions they encounter in the process of network technical services. For example, cooperate with network equipment manufacturers to obtain common problems they encounter in the process of equipment after-sales support and the corresponding repair and configuration solutions;

[0041] The data collected from the above three aspects are cleaned and preprocessed: (1) Remove noise data: The collected raw data may contain a large amount of noise information, such as advertisements, irrelevant replies, repeated content, etc., so these noise data need to be filtered out. For example, for data captured from online forums, if it is just simple replies without substantive content such as likes and thanks, they can be removed by keyword filtering or text length judgment; (2) Unify data format: Unify data from different sources into a standard format that the system can handle. For example, for problem descriptions and solutions, a unified text format is used, and the text is encoded and converted to ensure data consistency and compatibility; (3) Data classification processing: In order to facilitate subsequent keyword extraction, the collected data needs to be classified. For example, network technology problems can be classified according to different fields, such as network security, network architecture, wireless network, etc.

[0042] The main and secondary keyword extraction module uses appropriate word segmentation tools, such as Jieba for Chinese and NLTK for English, to split the problem description text into individual words or phrases. For example, for the Chinese question "What should I do if my wireless network always disconnects?", after using Jieba for word segmentation, we get {"我","的","无线网络","总是","断网","怎么办"};

[0043] Link adjacent word segments in the same sentence and set the initial weight of these links to 1. For example, after the above Jieba word segmentation, we form "我-的", "的-无线网络", "无线网络-总是", "总是-断网", "断网-怎么办", and their initial weights are all 1;

[0044] Perform the above processing on multiple network technology problem sentences, count the number of times each pair of linked word segments co - occur in all sentences, and use it as the weight of the edge to construct a co - occurrence graph of word segments. For example, if the link "网络-延迟" appears in 10 sentences, then its weight is updated to 10;

[0045] According to the co - occurrence graph, count the number of edges connected to each word segment, that is, the degree of the node. Here, the word segment can also be called a node in the graph. For example, the node "网络" is connected to 5 edges such as "如何-网络" and "网络-延迟", so its degree is 5;

[0046] Select the word segment corresponding to the node with the highest degree as the main keyword. If there are multiple nodes with the same and highest degree, according to specific business requirements, select one of them as the main keyword, or further screen by considering other factors, such as the importance of the word, etc.;

[0047] Starting from the node corresponding to the main keyword, traverse the graph structure in hierarchical order. First, visit the direct neighbor nodes of the main keyword, record their weights, and then visit the neighbor nodes of these neighbor nodes, and so on. During the traversal process, for each newly visited node, calculate the product of the weights of all edges on the path from the main keyword to this node as the comprehensive weight of this node. For example, if the edge weight from "网络服务器" to "操作系统" is 5, and the edge weight from "操作系统" to "内核参数" is 3, then the comprehensive weight of "内核参数" is 5×3 = 15;

[0048] As the path length increases, the relevance between the node and the main keyword may gradually weaken. Therefore, an attenuation factor ɑ is introduced to adjust the comprehensive weight. For example, each time an edge is passed, the weight is multiplied by a attenuation factor, such as ɑ = 0.9. The farther the node is from the main keyword, the lower its comprehensive weight will be, which is more in line with the characteristic that semantic relevance weakens with distance. For example, the comprehensive weight from "network server" to "operating system" and then to "kernel parameters" is 5×3×0.9=13.5.

[0049] Calculate the comprehensive weight Wi of all nodes directly and indirectly connected to the main keyword, i∈[1,n], where n represents the number of all nodes except the main keyword node. Set an appropriate threshold Wth. If Wi>Wth, the word segment corresponding to the i-th node is used as the secondary keyword. The setting of the threshold Wth needs to be adjusted according to the actual data and application scenarios. A too high threshold may lead to the omission of important secondary keywords, while a too low threshold may introduce too many irrelevant words.

[0050] Search heat calculation module, according to the formula Calculate the search popularity of the main keyword and a single sub-keyword, where F is the search frequency, that is, the number of times the main keyword and sub-keyword combination is searched within a specific time period. The more searches, the more attention the problem combination receives. R is the user feedback score, ranging from -100 to 100. It comprehensively considers the user's feedback on the search results, such as the user's satisfaction with the solution, the match between the search results and the user's needs, etc. If the user is very satisfied with the search results and the feedback is positive, then R is a positive value. If the user is dissatisfied with the search results and the feedback is negative, then R is negative. Negative values, R, enable dynamic adjustment of search popularity based on actual user experience; d is the search depth adjustment coefficient, ranging from [0, 1]. It is used to adjust the degree to which search popularity is affected by time decay. If d is close to 1, it means that time decay has a greater impact on search popularity. If d is close to 0, it means that time decay has a smaller impact. The value of d can be dynamically adjusted based on data characteristics and business needs. For example, for fields with rapid technological updates, d can be appropriately larger to highlight the popularity of new issues. For relatively stable fields, d can be smaller.

[0051] In addition to the search popularity of the combination of the main keyword and a single sub-keyword, the search popularity of the combination of the main keyword and multiple sub-keywords should also be considered. Since there may be semantic associations between multiple sub-keywords, and this association will affect their weight in the search popularity calculation, the specific operation steps are as follows: (1) Calculate the semantic similarity matrix: Use the word vector model to calculate the semantic similarity between each two sub-keywords. Assuming that there are m sub-keywords K1, K2, ..., Km, the calculated semantic similarity matrix M is an m×m matrix, where Mij represents the semantic similarity between sub-keywords Ki and Kj; (2) Calculate the overall semantic association degree: In order to obtain the overall semantic association degree of multiple sub-keywords, calculate the average value of all non-diagonal elements in the matrix, i≠j, and the specific formula is: S total The value range is [0,1]. The closer the value is to 1, the stronger the semantic association between the sub-keywords is. (3) Search heat adjustment: According to the formula S'=S×(1+β×(S total -γ)) to obtain the search popularity of the main keyword and multiple sub-keywords combined, where β is the weight adjustment coefficient, which is used to control the influence of semantic association on search popularity. It can be adjusted according to actual data and business needs. Generally, β>0, γ is the baseline value of semantic association. If S total >γ, increase the search popularity score, if S total <γ, reduce the search popularity score. The value of γ is usually around 0.5 and can be fine-tuned according to actual conditions;

[0052] When there are many secondary keywords, data sparsity may occur, that is, the search frequency of some secondary keyword combinations is extremely low or even zero, that is, F<λ, λ is a very small number, and the value range is usually [0,1]. The semantic similarity between secondary keywords can be used to estimate the search frequency: collect combinations Ct with F<λ, which contain main keywords and multiple secondary keywords, and screen out combinations Cr that are the same or similar to the main keywords in Ct, r∈[1,p], p is the total number of combinations, calculate the semantic similarity Sr between Ct and Cr according to the word vector model, find the combination Cr corresponding to max(Sr), the search frequency of Cr is Fr, and the estimated search frequency of Ct Festimate=max(Sr)×Fr.

[0053] The database storage module sorts network technology problems and their corresponding solutions according to the search popularity and stores them in the search popularity database. The database structure includes a primary keyword plus a secondary keyword field and a corresponding solution field.

[0054] User question matching module. When a user enters a network technology question, it extracts the main keywords and secondary keywords in the question, and counts the total number of characters Total of the extracted main keywords and secondary keywords. According to the main keywords, it locates a series of combinations related to the main keywords in the search popularity database, and extracts the combinations with a search popularity greater than Sth from them. It obtains the total number of characters Totali of the main keywords and secondary keywords in each combination, where i ∈ [1, e], and e is the number of combinations with a search popularity greater than Sth. If Totali <= Total, it adds the corresponding ith combination to the alternative comparison combinations. By comparing the character counts, it can preferentially exclude those combinations with significantly different character counts and less likely to match, narrowing the search scope and improving the matching efficiency. For example, if the user enters "Network fault repair method" with a total character count of 8, and in the database there is "Network fault" with a total character count of 4, combinations with a character count less than or equal to 8 will be included in the alternatives, while "Detailed settings steps for network security protection system" with a total character count of 14, combinations with a character count greater than 8 will be initially excluded.

[0055] For the main keywords and secondary keywords in the alternative comparison combinations, obtain the code points corresponding to each character according to the Unicode table, convert them into decimal numbers, and obtain the digital vector corresponding to each group of data. If Totali < Total, add 0 after the corresponding digital vector, and the number of added 0s is Total - Totali, and the transformed digital vector is Vcompi. For the main keywords and secondary keywords of the extracted user, also convert each character into the corresponding decimal number according to the Unicode table, and obtain the corresponding digital vector as Vuser. Using the uniqueness of Unicode encoding, convert the character information into digital form for subsequent calculation of the matching degree.

[0056] Calculate the matching degree Gi between the vector Vcompi and Vuser. If Gi > Gth, where Gth is the set matching degree threshold, add the solution to the corresponding network technology question in the search popularity database to the database to be displayed. Take out the solutions from the database to be displayed according to the size of the matching degree Gi and display them to the user one by one, enabling the user to first see the most relevant information.

[0057] Some of the data in the above formula are numerically calculated after removing their dimensions, and the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

[0058] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. The network technology service consultation intelligent matching system based on artificial intelligence is characterized by: It includes: A search popularity calculation module, which calculates the search popularity according to the search frequency of the combination of the main keyword and the secondary keyword, the user feedback score, the search depth adjustment coefficient, and the semantic similarity matrix; A database storage module, which sorts the network technology problems and their corresponding solutions according to the search popularity and stores them in the search popularity database. The database structure includes a main keyword plus a secondary keyword field and a corresponding solution field; A user problem matching module, which extracts the main keyword and the secondary keyword of the user input problem, counts the total number of characters Total of the two, locates the combination about the main keyword in the search popularity database according to the main keyword, extracts the combination with the search popularity greater than the search popularity threshold Sth, calculates the total number of characters Totali of the main keyword and the secondary keyword of this combination. If Totali <= Total, add this combination to the alternative comparison combinations, and obtain the corresponding digital vector according to the Unicode table for the combination of the main keyword and the secondary keyword. If Totali < Total, add 0 after the obtained digital vector, and the number of added 0s is Total - Totali. The obtained transformed digital vector is Vcompi. Also obtain the corresponding digital vector Vuser for the combination of the user's main keyword and secondary keyword according to the Unicode table, and calculate the matching degree Gi between the vector Vcompi and Vuser. If Gi > Gth, add the solution of the corresponding network technology problem in the search popularity database to the待展示数据库 (to be displayed database), and take out the solution from the待展示数据库 (to be displayed database) according to the size of the matching degree Gi and display it to the user. Here, Gth is the set matching degree threshold, i ∈ [1, e], and e is the number of combinations with the search popularity greater than Sth.

2. The network technology service consultation intelligent matching system based on artificial intelligence according to claim 1 is characterized in that: Before the search popularity calculation module, there are also a data collection module and a main and secondary keyword extraction module. The data collection module is used to collect historical data, and the historical data includes network technology problems and corresponding solutions; the main and secondary keyword extraction module is used to perform word segmentation on the network technology problems, construct a co-occurrence graph between the word segments, and extract the main keyword and the secondary keyword from the co-occurrence graph. The co-occurrence graph can reveal the relationship and weight between the keywords.

3. The network technology service consultation intelligent matching system based on artificial intelligence according to claim 2 is characterized in that: The specific steps for extracting the main keyword and the secondary keyword are as follows: First, use a word segmentation tool to split the problem description text into individual words or phrases; Second, link the adjacent word segments in the same sentence and set the initial weight of these links to 1; Perform the operations of the above steps 一 and 二 on multiple network technology problem sentences, count the number of times each pair of linked word segments co-occur in all sentences, and use it as the weight of the edge to construct a co-occurrence graph between the word segments; According to the co-occurrence graph, count the number of edges connected to each node, that is, the degree of the node, and the node represents a word segment; Select the word segment corresponding to the node with the highest degree as the main keyword. If there are multiple nodes with the same and highest degree, select one of them as the main keyword by yourself; Starting from the node corresponding to the main keyword, traverse the co-occurrence graph in hierarchical order. First, visit the direct neighbor nodes of the main keyword and record their weights. Then visit the neighbor nodes of these neighbor nodes, and so on. During the traversal process, for each newly visited node, calculate the product of the weights of all edges on the path from the main keyword to the node as the comprehensive weight of the node. The attenuation factor ɑ is introduced to adjust the comprehensive weight, and the comprehensive weight Wi of all nodes directly and indirectly connected to the main keyword is obtained, i∈[1,n], n represents the number of all other nodes except the main keyword node; Set a threshold Wth. If Wi>Wth, the word corresponding to the i-th node is used as the secondary keyword.

4. The network technology service consultation intelligent matching system based on artificial intelligence according to claim 1 is characterized in that: After combining the primary keyword and the secondary keyword to obtain the code point corresponding to each character according to the Unicode table, the code point needs to be converted into a decimal number.

5. The network technology service consultation intelligent matching system based on artificial intelligence according to claim 1 is characterized in that: The search popularity of the main keyword and a single sub-keyword is calculated using the following formula: ； Where F is the search frequency, R is the user feedback score, ranging from -100 to 100, T is the search time attenuation factor, and d is the search depth adjustment coefficient, ranging from 0 to 1.

6. The network technology service consultation intelligent matching system based on artificial intelligence according to claim 5 is characterized in that: The specific steps to calculate the search popularity of the main keyword and multiple sub-keywords are: The word vector model is used to calculate the semantic similarity between every two sub-keywords. Assuming there are m sub-keywords K1, K2, ..., Km, the calculated semantic similarity matrix M is an m×m matrix, where Mij represents the semantic similarity between sub-keywords Ki and Kj. Calculate the average value of all off-diagonal elements in the matrix. The specific formula is: , S total The value range of is [0,1], where i≠j; According to the formula S'=S×(1+β×(S total -γ)) obtains the search popularity of the combination of the main keyword and multiple sub-keywords, where β is the weight adjustment coefficient and γ is the baseline value of the semantic association degree.

7. The network technology service consultation intelligent matching system based on artificial intelligence according to claim 6 is characterized in that: When data sparsity occurs, the specific method of using the semantic similarity between sub-keywords to adjust the search frequency is as follows: Collect combinations Ct with F < λ, where Ct contains a main keyword and multiple sub-keywords, and filter out combinations Cr that are identical or similar to the main keyword of Ct, where r∈[1,p], p is the total number of combinations, and λ∈[0,1]; Calculate the semantic similarity Sr between Ct and Cr based on the word vector model, and find the combination Cr corresponding to max(Sr); The search frequency Fr of Cr is calculated, and the estimated search frequency Festimate of Ct is obtained: max(Sr)×Fr.

Images