Method and device for determining poor network quality optimization scheme, computer equipment, readable storage medium and program product

Abstract

The invention relates to a method and a device for determining a poor network quality optimization scheme, computer equipment, a readable storage medium and a program product. The method comprises the following steps: acquiring a poor network quality problem of a target area; querying a pre-constructed vector database according to the problem of poor network quality; information of a plurality of poor-quality cases is stored in the vector database; under the condition that the first matching result of the poor network quality problem is not queried from the vector database, querying a pre-constructed graph database according to the poor network quality problem to obtain a second matching result; the graph database stores a mapping relationship among poor quality phenomena, poor quality root causes, optimization schemes and occurrence times; and calling a second inference model to process the second matching result to obtain a target optimization scheme for the poor network quality problem of the target area. By adopting the method, the determination efficiency of the poor-quality optimization scheme can be improved.

Description

Technical Field

[0001] The present application relates to the field of big data technology, and in particular to a method, device, computer equipment, computer-readable storage medium and computer program product for determining a network quality optimization solution. Background Art

[0002] With the development of network technology, base stations are becoming more and more important. Currently, the number of base stations is increasing, and the difficulty of operation and maintenance is increasing. When there are some performance problems in the network, which affects user use, base station cell optimization, that is, quality optimization, is required.

[0003] In related technologies, poor wireless network quality is generally determined through the experience of network optimization experts to determine the root cause of the poor quality and propose an optimization plan. This method is highly dependent on the personal knowledge of experts, and the root cause of the poor quality must be determined before the optimization plan is determined. This step-by-step approach has low operation and maintenance efficiency. Summary of the invention

[0004] Based on this, it is necessary to provide a method, device, computer equipment, computer-readable storage medium and computer program product for determining a network quality poor optimization solution to address the technical problem of low operation and maintenance efficiency of the above-mentioned method of first determining the root cause of the quality poor and then determining the optimization solution.

[0005] In a first aspect, the present application provides a method for determining a network quality optimization solution. The method comprises:

[0006] Get the poor network quality problem in the target area;

[0007] According to the network poor quality problem, query a pre-built vector database; the vector database stores information of multiple poor quality cases;

[0008] In the case that the first matching result of the network poor quality problem is not queried from the vector database, querying a pre-built graph database according to the network poor quality problem to obtain a second matching result; the graph database stores a mapping relationship between poor quality phenomena, poor quality root causes, optimization solutions and occurrence times;

[0009] The second inference model is called to process the second matching result to obtain a target optimization solution for the poor network quality problem in the target area.

[0010] In one embodiment, the graph database is constructed in the following manner:

[0011] Obtain network data for the sample area;

[0012] Extracting knowledge from the network data to form a plurality of quality difference data combinations; each quality difference data combination includes a quality difference phenomenon, a quality difference root cause and an optimization solution;

[0013] The number of occurrences of each poor quality data combination is counted to obtain a mapping relationship between the poor quality phenomenon, the poor quality root cause, the optimization solution and the number of occurrences to form the graph database.

[0014] In one embodiment, querying a pre-built graph database according to the network quality problem to obtain a second matching result includes:

[0015] Performing vectorization processing on the network poor quality problem, and / or extracting keywords of the network poor quality problem;

[0016] Based on the network quality problem vector and / or keywords, the graph database is queried to obtain the second matching result.

[0017] In one embodiment, the vector database is constructed in the following manner:

[0018] Obtain multiple poor quality case documents;

[0019] Slice each poor quality case document separately, and vectorize each obtained slice;

[0020] The vector database is constructed based on the vectorized slices of each poor quality case document.

[0021] In one embodiment, querying a pre-built vector database according to the network quality problem includes:

[0022] Performing vector processing on the network quality problem;

[0023] Matching the network quality problem vector with each slice in the vector database;

[0024] A target slice whose matching degree with the network quality problem vector is greater than a threshold is selected from the slices as the first matching result.

[0025] In one embodiment, the method further comprises:

[0026] When a first matching result of the network poor quality problem is queried from the vector database, a first reasoning model is called to process the first matching result to obtain a target optimization solution for the network poor quality problem in the target area.

[0027] In a second aspect, the present application also provides a device for determining a network quality optimization solution. The device includes:

[0028] The acquisition module is used to obtain the network quality problem in the target area;

[0029] A first query module is used to query a pre-built vector database according to the network poor quality problem; the vector database stores information of multiple poor quality cases;

[0030] A second query module is used to query a pre-built graph database according to the network poor quality problem to obtain a second matching result when the first matching result of the network poor quality problem is not found from the vector database; the graph database stores a mapping relationship between poor quality phenomena, poor quality root causes, optimization solutions and occurrence times;

[0031] The reasoning module is used to call the second reasoning model to process the second matching result to obtain a target optimization solution for the network quality problem in the target area.

[0032] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0033] Get the poor network quality problem in the target area;

[0034] According to the network poor quality problem, query a pre-built vector database; the vector database stores information of multiple poor quality cases;

[0035] In the case that the first matching result of the network poor quality problem is not queried from the vector database, querying a pre-built graph database according to the network poor quality problem to obtain a second matching result; the graph database stores a mapping relationship between poor quality phenomena, poor quality root causes, optimization solutions and occurrence times;

[0036] The second inference model is called to process the second matching result to obtain a target optimization solution for the poor network quality problem in the target area.

[0037] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0038] Get the poor network quality problem in the target area;

[0039] According to the network poor quality problem, query a pre-built vector database; the vector database stores information of multiple poor quality cases;

[0040] In the case that the first matching result of the network poor quality problem is not queried from the vector database, querying a pre-built graph database according to the network poor quality problem to obtain a second matching result; the graph database stores a mapping relationship between poor quality phenomena, poor quality root causes, optimization solutions and occurrence times;

[0041] The second inference model is called to process the second matching result to obtain a target optimization solution for the poor network quality problem in the target area.

[0042] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0043] Get the poor network quality problem in the target area;

[0044] According to the network poor quality problem, query a pre-built vector database; the vector database stores information of multiple poor quality cases;

[0045] In the case that the first matching result of the network poor quality problem is not queried from the vector database, querying a pre-built graph database according to the network poor quality problem to obtain a second matching result; the graph database stores a mapping relationship between poor quality phenomena, poor quality root causes, optimization solutions and occurrence times;

[0046] The second inference model is called to process the second matching result to obtain a target optimization solution for the poor network quality problem in the target area.

[0047] The above-mentioned method, device, computer equipment, storage medium and computer program product for determining the network quality optimization solution first query the pre-constructed vector database according to the network quality problem of the target area; if the first matching result of the network quality problem is not found from the vector database, then query the pre-constructed graph database according to the network quality problem to obtain the second matching result; then call the second reasoning model to process the second matching result to obtain the target optimization solution for the network quality problem of the target area. This method pre-constructs a vector database, a graph database and a reasoning model, and uses them in combination. It only needs to input the network quality problem to obtain the optimization solution, without first determining the root cause of the quality difference, thereby improving the efficiency of determining the quality optimization solution; further, this method takes into account the efficient retrieval capability of the vector database and the deep reasoning capability of the graph database, and adopts the method of first querying the vector database, and then querying the graph database when the corresponding matching result is not found. The dual recall of vectors and knowledge graphs can effectively ensure the balance between the efficiency and accuracy of determining the quality optimization solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A flowchart of a method for determining a network quality optimization solution in one embodiment;

[0049] Figure 2 A schematic diagram of a flow chart of a map database construction step in one embodiment;

[0050] Figure 3 A schematic diagram of a process of constructing a vector database in one embodiment;

[0051] Figure 4 It is a flowchart of a method for determining a network quality optimization solution in another embodiment;

[0052] Figure 5 A schematic diagram of the architecture of a method for determining a network quality optimization solution in one embodiment;

[0053] Figure 6 It is a structural block diagram of a device for determining a network quality optimization solution in one embodiment;

[0054] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0056] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0057] In one embodiment, Figure 1 As shown, a method for determining a network quality optimization solution is provided. This embodiment uses the method applied to a terminal as an example. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. Among them, the terminal can be but is not limited to various personal computers, laptops, smart phones, tablets, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server can be implemented as an independent server or a server cluster consisting of multiple servers. In this embodiment, the method includes the following steps:

[0058] Step S110, obtaining the network quality problem of the target area.

[0059] The target area indicates the area where the network quality optimization solution needs to be determined. In practical applications, the target area may be a base station cell.

[0060] Among them, the network quality problem can be a description of the poor quality situation of the target area, such as unstable connection, low data rate, etc.; in some embodiments, it can also be the original network quality data of the target area, such as signal strength, data rate, delay, etc.

[0061] In a specific implementation, the content input by the user may be received, and the problem of poor network quality may be obtained according to the content input by the user.

[0062] Step S120, according to the network poor quality problem, query the pre-built vector database; the vector database stores information of multiple poor quality cases.

[0063] In the specific implementation, after obtaining the network poor quality problem, you can first query the vector database to find similar poor quality cases to obtain an optimization solution. Specifically, before querying, the obtained network poor quality problem must be vectorized and converted into a vector form. For example, a pre-trained vectorization model can be used to convert the network poor quality problem from text form into a numerical vector suitable for fast similarity comparison. After completing the conversion, a similarity search is performed in the vector database to find cases that are close to the current network poor quality problem vector. If a highly related poor quality case is found, it is recorded as the first matching result.

[0064] Step S130, when the first matching result of the network poor quality problem is not found from the vector database, a pre-built graph database is queried according to the network poor quality problem to obtain a second matching result; the graph database stores the mapping relationship between the poor quality phenomenon, the root cause of the poor quality, the optimization solution and the number of occurrences.

[0065] In the specific implementation, if the first matching result that meets the conditions cannot be found in the vector database, for example, a threshold can be set. If the similarity between the content retrieved in the vector database and the network quality problem is lower than the threshold, it is considered that no useful first matching result has been found, and the graph database will continue to be queried. The graph database pre-stores the mapping relationship between various quality problems, quality root causes, optimization solutions, and occurrence times. This information is organized into a graph structure, which enables efficient relationship query and reasoning.

[0066] When querying the graph database, based on the network quality problem, a query request is sent to the graph database to find the quality difference phenomenon related to the problem. Based on the determined quality difference phenomenon, the corresponding quality difference root cause, optimization plan and number of occurrences are obtained as the second matching result.

[0067] It can be understood that there may be multiple poor quality phenomena related to the poor network quality problem obtained by querying the graph database, so multiple groups of matching results will be obtained: <poor quality phenomenon, poor quality root cause, optimization plan, number of occurrences>.

[0068] Step S140: calling the second reasoning model to process the second matching result to obtain a target optimization solution for the poor network quality problem in the target area.

[0069] In the specific implementation, the second inference model is an inference model constructed for the matching results of the graph database. The second inference model can be a large model, for example, a general large model or a large model fine-tuned with network optimization corpus. The model can derive a specific optimization solution for the target area by analyzing the association between the root cause of poor quality and the optimization solution. The obtained second matching results usually include the corresponding poor quality phenomenon, root cause, optimization solution and its occurrence frequency. The second inference model is used to analyze and process these data to generate a target optimization solution for the target area.

[0070] It can be understood that the second matching result may include multiple situations. Therefore, in the reasoning process of the second reasoning model, the model will also consider the number of occurrences as a reasoning basis for the target optimization plan. For example, the more times a certain quality difference phenomenon occurs, the greater the probability that the optimization plan corresponding to the quality difference phenomenon will be selected into the target optimization plan.

[0071] In some embodiments, considering that most quality difference optimization solutions are in a unified format and have a low degree of humanization, a wireless expert quality difference optimization solution output template can be defined. The second reasoning model is preset as a network optimization expert role and given a certain information summarization capability. The second reasoning model is required to output results in a specified format and provide generated samples for few-shot learning, so that the generated effect is more stable.

[0072] In this step, a multi-thinking chain reasoning method is adopted to match different thinking chain capabilities according to the different types of knowledge recall. For example, for a graph database, reasoning is performed based on the large model knowledge graph reasoning and the number of occurrences of the quality difference optimization solution, so that the returned knowledge is closer to the current network world.

[0073] In the above-mentioned method for determining the network quality optimization solution, according to the network quality problem in the target area, the pre-constructed vector database is first queried; when the first matching result of the network quality problem is not found from the vector database, the pre-constructed graph database is queried according to the network quality problem to obtain the second matching result; then the second reasoning model is called to process the second matching result to obtain the target optimization solution for the network quality problem in the target area. This method pre-constructs a vector database, a graph database and an inference model and uses them in combination. It only needs to input the network quality problem to obtain the optimization solution, without first determining the root cause of the quality difference, thereby improving the efficiency of determining the quality difference optimization solution; further, this method takes into account the efficient retrieval capability of the vector database and the deep reasoning capability of the graph database, and adopts the method of first querying the vector database, and then querying the graph database when no corresponding matching result is found. The dual recall of vectors and knowledge graphs can effectively ensure the balance between the efficiency and accuracy of determining the quality difference optimization solution.

[0074] In an exemplary embodiment, if Figure 2 As shown, the graph database in the above step S130 is constructed in the following way:

[0075] Step S210, obtaining network data of the sample area.

[0076] Specifically, through the big data platform, the user measurement report (MR) data, base station cell performance report (KPI) data, alarm data, engineering parameter data, as well as expert cases and work order return information in the sample area are obtained as network data. Specifically including:

[0077] (1) Based on the cell dimension, summarize MR data, KPI data, alarm data, and engineering parameter data.

[0078] (2) Taking the community as the dimension, summarize all the historical poor quality communities and extract the work order data (including return order data) from the work order system.

[0079] (3) Taking the cell as the dimension, extract the MR data, KPI data, alarm data, and engineering parameter data when poor quality occurs, and associate them with the list of cells with poor quality to form poor quality phenomenon data mainly based on the type of poor quality and the data of poor quality.

[0080] (4) Based on the community dimension, correlate the quality difference phenomenon data and work order return data.

[0081] Step S220, extracting knowledge from the network data to form a plurality of quality difference data combinations; each quality difference data combination includes a quality difference phenomenon, a quality difference root cause and an optimization solution.

[0082] In the specific implementation, the knowledge graph of the big model can be used to prompt the engineering to extract knowledge. Specifically, the big model can be given certain roles, such as knowledge graph engineering experts and wireless network optimization experts, who are good at accurately extracting entities (subjects, objects) and relationships of the wireless network knowledge graph from the text, and can make appropriate summary descriptions of the meanings of entities and relationships; and the big model is given certain capabilities, such as wireless network entity extraction capabilities, wireless network relationship extraction capabilities, and wireless network context association capabilities; and the big model is given certain constraints, such as using as much information in the associated context as possible to enrich the content of entities and relationships, not making up at will, etc.; and the big model data format is agreed upon, such as <quality difference phenomenon-quality difference root cause-quality difference optimization plan>. In this way, multiple quality difference data combinations including quality difference phenomenon, quality difference root cause and optimization plan are obtained.

[0083] Step S230, count the number of occurrences of each poor quality data combination, obtain the mapping relationship between the poor quality phenomenon, the poor quality root cause, the optimization solution and the number of occurrences, and form a graph database.

[0084] In the specific implementation, after obtaining the quality difference data combination, the number of occurrences of each quality difference data combination can be counted. For example, each time knowledge extraction is performed, the obtained quality difference data combination can be compared with the existing quality difference data combination, and the similarity can be judged. If the similarity meets the conditions (for example, greater than the threshold), it is regarded as the same quality difference data combination, and the number of occurrences of the quality difference data combination is increased by 1; if the similarity does not meet the conditions, it is regarded as a new quality difference data combination, and its number of occurrences is recorded as 1, and so on. After all data processing is completed, a four-tuple of <quality difference phenomenon-quality difference root cause-optimization plan-number of occurrences> can be formed and stored in the graph database.

[0085] In this embodiment, semi-structured data such as work orders and cases are collected, and wireless entity extraction and wireless relationship extraction are performed through a large model. At the same time, similar knowledge is merged and counted to form a four-tuple of <poor quality phenomenon-poor quality root cause-poor quality optimization plan-number of occurrences>, and stored in a graph database to realize the construction of a wireless quality difference knowledge graph, so that the database can be directly searched in the future to determine the optimization plan for the network quality difference problem, thereby improving the efficiency of determining the network quality difference optimization plan.

[0086] In an exemplary embodiment, the above-mentioned step S130 queries a pre-constructed graph database according to the poor network quality problem to obtain a second matching result, including: vectorizing the poor network quality problem, and / or extracting keywords of the poor network quality problem; based on the poor network quality problem vector, and / or keywords, querying the graph database to obtain a second matching result.

[0087] In the specific implementation, the pre-built graph database can be queried based on the network quality problem through vector search, keyword search, or vector search and keyword mixed search. Specifically, after the user enters the network quality problem, a search formula can be constructed based on the network quality problem, and the similarity between the search formula and each record in the database can be calculated using methods such as cosine similarity and Euclidean distance. According to the similarity score, several records that best match the query are returned, and can be arranged in descending order according to the similarity.

[0088] More specifically, the text of the network quality problem can be directly converted into a vector as the first search formula; based on the similarity between the first search formula and each record in the graph database, the second matching result is obtained. It is also possible to extract keywords in the network quality problem through natural language technology to construct a second search formula; based on the similarity between the second search formula and each record in the graph database, the second matching result is obtained. It is also possible to merge the matching results obtained by the first search formula with the matching results obtained by the second search formula as the final second matching result.

[0089] For example, obtain the result list and similarity obtained by the first search formula, and obtain the result list obtained by the second search formula, and evaluate the relevance of each result to the second search formula. Finally, merge the two results. For example, weights can be assigned to the two search methods. For each search result, the weights of the two search methods can be used to weight the sum of the results of the two search methods to obtain a total score. Then, sort the results according to the total score, and select the top results as the second matching result. For example, if the weight of the search result corresponding to the first search formula is α, and the weight of the search result corresponding to the second search formula is β, then the total score can be expressed as: α×similarity+β×relevance score.

[0090] In this embodiment, query methods such as vector search, keyword search, and a hybrid search of the two can complement each other and jointly improve the efficiency and accuracy of knowledge recall.

[0091] In an exemplary embodiment, if Figure 3 As shown, the above vector database is constructed in the following way:

[0092] Step S310, obtaining a plurality of poor quality case documents;

[0093] Step S320, slicing each poor quality case document, and vectorizing each obtained slice;

[0094] Step S330, constructing a vector database based on the vectorized slices of each poor quality case document.

[0095] In a specific implementation, case documents about poor network quality can be collected from various channels, and the slice length can be determined according to the content and structure of the document. Each poor quality case document is sliced ​​according to the slice length, so that each document can obtain multiple slices. Then, each obtained slice is vectorized through a pre-trained vectorization model, that is, the text slice is converted into a vector representation, and the vectorized slices of each poor quality case document are stored in a preset database to form a vector database.

[0096] In some embodiments, after obtaining the case document, semantic annotation can also be performed. For example, key elements such as quality difference phenomenon, root cause, optimization plan, etc. can be annotated in the case document to facilitate subsequent analysis.

[0097] In some embodiments, in order to increase contextual information, when slicing the poor quality case documents, overlapping slices (for example, each slice overlaps by 50%) may be set to capture more semantic associations.

[0098] In this embodiment, slices of case documents are stored in the form of high-dimensional vectors through vectorization, so that subsequent knowledge retrieval can be efficiently performed to obtain cases and optimization solutions similar to the user input, greatly reducing the time cost of traditional text retrieval and improving retrieval efficiency.

[0099] In an exemplary embodiment, in the above step S120, a pre-constructed vector database is queried based on the network quality problem, including: vectorizing the network quality problem; matching the network quality problem vector with each slice in the vector database; and selecting a target slice from each slice whose matching degree with the network quality problem vector is greater than a threshold value as the first matching result.

[0100] In the specific implementation, when querying the vector database, the network quality problem is first converted into a high-dimensional vector representation. Then, the network quality problem vector is matched with the vectors of all slices in the vector database by calculating the similarity (such as cosine similarity). A similarity threshold (such as 0.75) can be pre-defined, and from the matched slices, the target slices whose similarity with the network quality problem vector is higher than the threshold are screened out, and the screened target slices are returned as the first matching result.

[0101] In this embodiment, through vectorization and similarity retrieval, the query process is fast and efficient, and the response speed is greatly improved compared to traditional text retrieval.

[0102] In an exemplary embodiment, the method further includes: when a first matching result of the network poor quality problem is queried from the vector database, calling a first reasoning model to process the first matching result to obtain a target optimization solution for the network poor quality problem in the target area.

[0103] In a specific implementation, if the first matching result of the network quality problem is queried in the vector database, the first reasoning model is called to process the first matching result. The first reasoning model is a reasoning model constructed for the matching result of the vector database. The first reasoning model can also be a large model, for example, a general large model or a large model that has been fine-tuned with the network optimization corpus. The large model is preset as a network optimization expert role and is given certain skills and constraints. The large model is required to think about the topN possible root causes of quality differences and optimization solutions based on the matching results returned by the vector database, and there is a corresponding relationship between the root causes of quality differences and the optimization solutions.

[0104] In some embodiments, considering that most quality difference optimization solutions are in a unified format and have a low degree of humanization, a wireless expert quality difference optimization solution output template can be defined. The first reasoning model is preset as a network optimization expert role and given a certain information summarization capability. The first reasoning model is required to output results in a specified format and provide generated samples for few-shot learning, so that the generated effect is more stable.

[0105] In this embodiment, a multi-thinking chain reasoning method is adopted to match different thinking chain capabilities according to the different types of knowledge recall. For example, if it is a vector database, a large model that matches the recall results of the vector database is trained to perform knowledge graph reasoning, thereby improving the accuracy of the target optimization plan of reasoning.

[0106] refer to Figure 4 , is a flow chart of a method for determining a network quality difference optimization solution proposed in another embodiment. In this embodiment, the following steps are included:

[0107] Step S410: obtaining the network quality problem of the target area.

[0108] Step S420, vectorize the network quality problem, and match the obtained network quality problem vector with each slice in the vector database.

[0109] Step S430: If there is a target slice in each slice whose matching degree with the network quality problem vector is greater than a threshold, the target slice is taken as the first matching result.

[0110] Step S440: calling the first inference model to process the first matching result to obtain a target optimization solution for the poor network quality problem in the target area.

[0111] Step S450: If there is no target slice in each slice whose matching degree with the network quality problem vector is greater than a threshold, vectorization is performed on the network quality problem, and / or keywords of the network quality problem are extracted.

[0112] Step S460, based on the network quality problem vector and / or the keyword, query the graph database to obtain a second matching result.

[0113] Step S470: Call the second reasoning model to process the second matching result to obtain a target optimization solution for the poor network quality problem in the target area.

[0114] This method pre-builds a vector database, a graph database and an inference model, and uses them in combination. It only needs to input the network quality problem to obtain the optimization plan, without first determining the root cause of the poor quality, thereby improving the efficiency of determining the poor quality optimization plan; further, this method takes into account the efficient retrieval capability of the vector database and the deep reasoning capability of the graph database, and first queries the vector database, and then queries the graph database when no corresponding matching results are found. The dual recall method of vectors and knowledge graphs can effectively ensure the balance between the efficiency and accuracy of determining the poor quality optimization plan.

[0115] In one embodiment, in order to facilitate those skilled in the art to understand the embodiments of the present application, the following will be described with reference to specific examples of the accompanying drawings. Figure 5 , shows a schematic diagram of the architecture of a method for determining a network quality difference optimization solution, including a data processing module, a knowledge recall module, a reasoning module and a generation module. The process of determining a network quality difference optimization solution through each module includes:

[0116] (1) Collecting raw data: Through the big data platform, obtain user measurement report (MR) data, base station cell performance report (KPI) data, alarm data, engineering parameter data, expert cases and work order return information in the sample area.

[0117] (2) Knowledge extraction and storage: On the one hand, the large model knowledge graph is used to prompt the engineering department to extract knowledge, count the number of similar knowledge, and form a four-tuple: <phenomenon description-root cause of quality difference-optimization solution-number of occurrences>, which is then stored in the graph database; on the other hand, the case documents are sliced ​​and vectorized and stored in the vector database. Knowledge is stored in two different ways.

[0118] (3) Vector database recall: Since vector retrieval is faster, vector database recall is given priority. The vector database is queried based on the user's poor network quality problem. If a matching result that meets the threshold of the user's problem is found, the result list is returned (such as selecting the topN results) and the reasoning of step (5) is entered. If not, step (4) is entered to recall the graph database.

[0119] (4) Graph database recall: Based on the user’s poor network quality problem, relevant knowledge items are queried based on vector / keyword / hybrid retrieval and a list of results is returned.

[0120] In some embodiments, taking into account the high overhead of graph databases and combining the fast and slow thinking patterns of humans, after obtaining the user's poor network quality problem, a judgment can be made to determine the type of problem. If it is a simple problem, vector database retrieval is used first, and if it is a complex problem, graph database retrieval is used.

[0121] (5) Multi-thinking chain reasoning: Based on a general large model or a large model fine-tuned with network optimization corpus, and preset as a network optimization role. If the recall result is the result returned by the vector database, then the vector database thinking chain reasoning ability is constructed, allowing the large model to deduce the relevant quality difference optimization plan step by step based on the returned knowledge; if the recall result is the result returned by the graph database, then the knowledge graph thinking chain reasoning ability is constructed, allowing the large model to deduce the relevant quality difference optimization plan based on the graph reasoning content and the number of occurrences.

[0122] (5.1) Vector database reasoning: The big model is preset as a network optimization expert and is given certain skills and constraints. The big model is required to think about the top N possible root causes of poor quality and optimization solutions based on the results returned by the vector database, and there needs to be a corresponding relationship between the root causes of poor quality and the optimization solutions.

[0123] (5.2) Knowledge graph thinking chain reasoning: The big model is preset as a network optimization expert role and is given certain skills and constraints. The big model is required to think about the top N possible root causes of poor quality and optimization solutions based on the results returned by the graph database and the number of occurrences. There needs to be a corresponding relationship between the root causes of poor quality and the optimization solutions.

[0124] In this step, different thought chain reasoning is performed according to the different sources of the recalled knowledge, giving full play to the advantages of the two different data formats and making the reasoning results more accurate.

[0125] (6) Prompt engineering answer generation: Through few-shot (zero sample) prompt engineering, the large model outputs the final quality difference optimization plan in the specified format of the current quality difference phenomenon, reference work order phenomenon, quality difference root cause, and quality difference optimization plan, making the generated effect more stable.

[0126] In this step, based on the needs of front-line operation and maintenance personnel, an optimization plan that meets the needs of front-line personnel is generated through a preset optimization plan template.

[0127] In response to the current problems of wireless quality difference optimization, such as high reliance on expert knowledge, low automation rate, and great optimization difficulty, this application proposes a quality difference optimization solution based on a large model. On the one hand, it uses the large model to prompt the engineering to extract knowledge, count the number of similar knowledge, and form a four-tuple of <phenomenon description-quality difference root cause-optimization solution-number of occurrences> to form a graph database; on the other hand, it forms a vector database by slicing and vectorizing case documents; and stores knowledge in two different ways. During retrieval, multi-way retrieval is performed, and knowledge selection and aggregation are performed based on similarity. Finally, the large model prompts the engineering to output the results. Compared with the prior art, this application has the following beneficial effects:

[0128] 1. High efficiency: Taking advantage of the strong generalization ability of large models, using prompt engineering to automatically extract knowledge through large models, build knowledge graphs, and reduce manual operations.

[0129] 2. High accuracy: A vector database and a quadruple graph database are built, and the powerful reasoning ability of the large model is used to improve the accuracy.

[0130] 3. The dual recall method of vector and knowledge graph can effectively ensure the balance between quality optimization efficiency and accuracy.

[0131] 4. Highly user-friendly: By utilizing the generation capability of large models, root cause location and optimization solutions can be generated in a specified format in one step, which is closer to the needs of front-line optimization personnel.

[0132] 5. Strong sustainability: With the continuous iteration of knowledge, more knowledge will be automatically extracted and the recommendations will be more accurate.

[0133] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0134] Based on the same inventive concept, the embodiment of the present application also provides a device for determining a network quality difference optimization solution for implementing the method for determining the network quality difference optimization solution involved above. The implementation solution for solving the problem provided by the device is similar to the implementation solution recorded in the above method, so the specific limitations in the embodiments of the device for determining one or more network quality difference optimization solutions provided below can refer to the limitations of the method for determining the network quality difference optimization solution above, and will not be repeated here.

[0135] In one embodiment, Figure 6 As shown, a device for determining a network quality difference optimization solution is provided, comprising:

[0136] An acquisition module 610 is used to acquire the network quality problem of the target area;

[0137] The first query module 620 is used to query a pre-built vector database according to the network quality problem; the vector database stores information of multiple quality problems;

[0138] The second query module 630 is used to query a pre-built graph database according to the network poor quality problem to obtain a second matching result when the first matching result of the network poor quality problem is not found from the vector database; the graph database stores a mapping relationship between the poor quality phenomenon, the poor quality root cause, the optimization solution and the number of occurrences;

[0139] The reasoning module 640 is used to call the second reasoning model to process the second matching result to obtain a target optimization solution for the network quality problem in the target area.

[0140] In one of the embodiments, the device also includes a graph database construction module, which is used to obtain network data of the sample area; perform knowledge extraction on the network data to form multiple quality difference data combinations; each quality difference data combination includes a quality difference phenomenon, a quality difference root cause, and an optimization plan; count the number of occurrences of each quality difference data combination to obtain a mapping relationship between the quality difference phenomenon, the quality difference root cause, the optimization plan, and the number of occurrences to form a graph database.

[0141] In one embodiment, the second query module 630 is further used to vectorize the network poor quality problem and / or extract keywords of the network poor quality problem; based on the network poor quality problem vector and / or keywords, query the graph database to obtain a second matching result.

[0142] In one of the embodiments, the device also includes a vector database construction module, which is used to obtain multiple poor quality case documents; slice each poor quality case document separately, and vectorize each obtained slice; and construct a vector database based on the vectorized slices of each poor quality case document.

[0143] In one of the embodiments, the first query module 620 is also used to vectorize the network quality problem; match the network quality problem vector with each slice in the vector database; and filter out the target slice whose matching degree with the network quality problem vector is greater than a threshold from each slice as the first matching result.

[0144] In one embodiment, the reasoning module 640 is also used to call the first reasoning model to process the first matching result when a first matching result of the network quality problem is queried from the vector database, so as to obtain a target optimization solution for the network quality problem in the target area.

[0145] Each module in the above-mentioned device for determining the network quality optimization scheme can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each of the above modules.

[0146] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for determining a network quality optimization solution is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0147] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0148] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0149] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0150] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0151] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0152] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0153] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0154] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for determining a network quality optimization solution, characterized in that: The method comprises: Get the poor network quality problem in the target area; According to the network poor quality problem, query a pre-built vector database; the vector database stores information of multiple poor quality cases; In the case that the first matching result of the network poor quality problem is not queried from the vector database, querying a pre-built graph database according to the network poor quality problem to obtain a second matching result; the graph database stores a mapping relationship between poor quality phenomena, poor quality root causes, optimization solutions and occurrence times; The second inference model is called to process the second matching result to obtain a target optimization solution for the poor network quality problem in the target area.

2. The method according to claim 1, characterized in that The graph database is constructed in the following way: Obtain network data for the sample area; Extracting knowledge from the network data to form a plurality of quality difference data combinations; each quality difference data combination includes a quality difference phenomenon, a quality difference root cause and an optimization solution; The number of occurrences of each poor quality data combination is counted to obtain a mapping relationship between the poor quality phenomenon, the poor quality root cause, the optimization solution and the number of occurrences to form the graph database.

3. The method according to claim 2, characterized in that According to the network quality problem, querying a pre-built graph database to obtain a second matching result includes: Performing vectorization processing on the network poor quality problem, and / or extracting keywords of the network poor quality problem; Based on the network quality problem vector and / or the keyword, the graph database is queried to obtain the second matching result.

4. The method according to claim 1, characterized in that: The vector database is constructed in the following way: Obtain multiple poor quality case documents; Slice each poor quality case document separately, and vectorize each obtained slice; The vector database is constructed based on the vectorized slices of each poor quality case document.

5. The method according to claim 4, characterized in that The querying of a pre-built vector database according to the network quality problem includes: Performing vector processing on the network quality problem; Matching the network quality problem vector with each slice in the vector database; A target slice whose matching degree with the network quality problem vector is greater than a threshold is selected from the slices as the first matching result.

6. The method according to claim 1, characterized in that The method further comprises: When a first matching result of the network poor quality problem is queried from the vector database, a first reasoning model is called to process the first matching result to obtain a target optimization solution for the network poor quality problem in the target area.

7. A device for determining a network quality optimization solution, characterized in that: The device comprises: The acquisition module is used to obtain the network quality problem in the target area; A first query module is used to query a pre-built vector database according to the network poor quality problem; the vector database stores information of multiple poor quality cases; A second query module is used to query a pre-built graph database according to the network poor quality problem to obtain a second matching result when the first matching result of the network poor quality problem is not found from the vector database; the graph database stores a mapping relationship between poor quality phenomena, poor quality root causes, optimization solutions and occurrence times; The reasoning module is used to call the second reasoning model to process the second matching result to obtain a target optimization solution for the network quality problem in the target area.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for determining the network quality difference optimization solution according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for determining a network quality difference optimization solution according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method for determining a network quality difference optimization solution according to any one of claims 1 to 6 are implemented.

Images