Inspection report generation method and device for cloud network networking scene, medium and product

Abstract

The embodiment of the invention provides an inspection report generation method and device for a cloud network networking scene, a medium and a product, and relates to the technical field of computers, and the method comprises the steps: retrieving a document related to the semantics of abnormal item information from a predetermined knowledge base based on the abnormal item information, and taking the document as the context information of the abnormal item information; the abnormal item information is used for indicating an abnormal detection result of any inspection item of a detection object, and the detection object is one or more products used for cloud network networking; based on the context information and the exception item information, performing exception summary on the inspection items by using a predetermined large language model to obtain an exception item summary text; and generating an inspection report by using the abnormal item summary text. In the scheme, the context information can enhance the semantic understanding of the large language model for the abnormal item information, the abnormal item summary text generated by the large language model can more accurately describe the abnormal item information, the accuracy and readability of the inspection report are improved, and the interpretation difficulty of the inspection report is reduced.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device, medium and product for generating an inspection report in a cloud network networking scenario. Background Art

[0002] The network of cloud network customers is usually distributed in multiple regions and is built by a combination of cloud network products such as cloud enterprise network, load balancing, and public network elastic network address (Elastic Internet Protocol Address, EIP). This dynamic and complex cloud network architecture not only increases the flexibility and scalability of the network, but also brings challenges in management and maintenance. In this network environment, traditional inspection reports can only simply list the abnormal items found during the inspection process. The abnormal item list information is poorly readable and not easy to understand, making it difficult to interpret the report. Summary of the invention

[0003] The embodiments of the present application provide a method, device, medium and product for generating an inspection report in a cloud network networking scenario to alleviate or solve one or more technical problems existing in the prior art.

[0004] In a first aspect, an embodiment of the present application provides a method for generating an inspection report for a cloud network networking scenario, the method comprising: based on the acquired abnormal item information, retrieving documents semantically related to the abnormal item information from a predetermined knowledge base as context information of the abnormal item information; the abnormal item information is used to indicate the abnormal detection result of any inspection item of the detection object, and the detection object is one or more products used for cloud network networking; based on the context information and the abnormal item information, using a predetermined large language model to perform an abnormal summary on the inspection items to obtain an abnormal item summary text; and using the abnormal item summary text to generate an inspection report.

[0005] In a second aspect, an embodiment of the present application provides a method for generating an inspection report for a cloud network networking product, the method comprising: obtaining an abnormality detection result of any inspection item from the inspection result of a cloud network networking product; generating abnormal item information based on the abnormality detection result; based on the abnormal item information, retrieving documents semantically related to the abnormal item information from a predetermined knowledge base as context information of the abnormal item information; based on the context information and the abnormal item information, using a predetermined large language model to perform an abnormal summary of the inspection item to obtain an abnormal item summary text; and using the abnormal item summary text to generate an inspection report for the cloud network networking product.

[0006] In a third aspect, an embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory, and the processor implements any method of the embodiment of the present application when executing the computer program.

[0007] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the method of any one of the embodiments of the present application is implemented.

[0008] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which implements any method of the embodiments of the present application when executed by a processor.

[0009] According to the method of the embodiment of the present application, one or more products of the cloud network group are taken as detection objects. For the abnormal item information of the detection object, the semantically related documents retrieved from the knowledge base can be used as the context information of the abnormal item information, and the abnormal item information and its context information are abnormally summarized using the large language model, and the obtained abnormal item summary text is used to update the inspection report. In this method, by using the documents semantically related to the abnormal item information retrieved from the knowledge base as context information, the semantic understanding of the abnormal item information by the large language model can be better enhanced. On the basis of enhancing the semantic understanding, the abnormal item summary text generated by the large language model helps to more accurately describe the abnormal item information, which is beneficial to improve the accuracy and readability of the inspection report, reduce the difficulty of interpreting the inspection report, make the content of the inspection report easier to understand, and thus improve the quality of the inspection report.

[0010] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the multiple drawings represent the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments according to the present application and should not be regarded as limiting the scope of the present application.

[0012] Figure 1 A flow chart of a method for generating an inspection report for a cloud network networking scenario according to an embodiment of the present application is shown.

[0013] Figure 2 A flow chart of a method for generating an inspection report for a cloud network provided in an embodiment of the present application is shown.

[0014] Figure 3 A flow chart of a method for generating an inspection report for a cloud network networking according to an exemplary embodiment of the present application is shown.

[0015] Figure 4 A structural diagram of an inspection report generating device for a cloud network networking scenario provided in an embodiment of the present application is shown.

[0016] Figure 5 A structural diagram of an inspection report generating device for cloud network networking provided in an embodiment of the present application is shown.

[0017] Figure 6 A block diagram of an electronic device provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0018] In the following, only some exemplary embodiments are briefly described. As those skilled in the art will appreciate, the described embodiments may be modified in various ways without departing from the concept or scope of the present application. Therefore, the drawings and descriptions are considered to be exemplary in nature and not restrictive.

[0019] To facilitate understanding of the technical solutions of the embodiments of the present application, the following describes the related technologies of the embodiments of the present application. The following related technologies can be combined with the technical solutions of the embodiments of the present application as optional solutions, and they all belong to the protection scope of the embodiments of the present application.

[0020] In actual scenarios, cloud network inspection refers to the process of regular inspection and maintenance of one or more products in the cloud network. The products in the cloud network (hereinafter referred to as cloud network products) include but are not limited to at least one of the following: cloud enterprise network, load balancing, public network elastic network address (Elastic Internet Protocol Address, EIP), etc. Inspection helps to ensure the reliability, security and performance of cloud network products, and promptly discover and solve potential problems to avoid larger failures.

[0021] Since the cloud network architecture is dynamic and complex, this architecture feature increases the flexibility and scalability of the network, but also brings challenges in management and maintenance. In the complex and dynamic cloud network environment, the current cloud network inspection mainly tests the pre-written inspection items one by one, and lists the inspection results in the inspection report. The inspection report has poor readability, low reading efficiency, and is not easy to understand. It is difficult to interpret the report manually. In addition, the inspection report usually does not provide specific solutions for abnormal item information. Users lack an overall analysis perspective and find it difficult to locate the root cause of the inspection anomaly. If the report is interpreted and solutions are provided manually, the reading cost is high and professional knowledge is required. It takes a lot of time and manpower, and it is easy to miss key issues due to manual negligence and lack of experience.

[0022] The embodiments of the present application aim to provide a method for generating an inspection report in a cloud network networking scenario, an electronic device, a storage medium, and a program product, which are used to reduce the difficulty of interpreting the report, make the report easier to understand, and improve the readability of the report.

[0023] It should be noted that the above application scenarios or application examples provided in the embodiments of the present application are for ease of understanding, and the embodiments of the present application do not specifically limit the application of the technical solution. In addition, 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 need to comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0024] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The several specific embodiments listed can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0025] Figure 1 A flowchart of a method for generating an inspection report in a cloud network networking scenario according to an embodiment of the present application is shown. Figure 1 As shown, the method may include the following steps S110-S130.

[0026] S110, based on the acquired abnormal item information, retrieve documents semantically related to the abnormal item information from a predetermined knowledge base as context information of the abnormal item information; the abnormal item information is used to indicate the abnormal detection result of any inspection item of the detection object, and the detection object is one or more products used for cloud network networking.

[0027] S120, based on the context information and the abnormal item information, using a predetermined large language model to perform abnormal summary on the inspection items to obtain an abnormal item summary text.

[0028] S130, generating an inspection report using the abnormal item summary text.

[0029] According to the method of the embodiment of the present application, one or more products of the cloud network networking are taken as detection objects. For the abnormal item information of the detection object, the semantically related documents retrieved from the knowledge base can be used as the context information of the abnormal item information, and the abnormal item information and its context information are abnormally summarized using the large language model, and the obtained abnormal item summary text is used to generate an inspection report. In this method, by using the documents semantically related to the abnormal item information retrieved from the knowledge base as context information, the semantic understanding of the abnormal item information by the large language model can be better enhanced. On the basis of enhancing the semantic understanding, the abnormal item summary text generated by the large language model helps to more accurately describe the abnormal item information, which is conducive to improving the accuracy and readability of the inspection report, reducing the difficulty of interpreting the inspection report, making the content of the inspection report easier to understand, and thus improving the quality of the inspection report.

[0030] In some embodiments, the abnormal item information of the detection object may include multiple acquisition methods. The first acquisition method is that after detecting each inspection item of the cloud network networking product to obtain the detection results of each inspection item, the abnormal detection results can be directly obtained from the detection results of each inspection item, and the abnormal item information is generated according to the abnormal detection results. The second acquisition method is that if the inspection process in the actual scenario includes generating an initial inspection report based on the detection results of each inspection item, the abnormal detection results can be obtained from the initial inspection report, and then the abnormal item information is generated according to the acquired abnormal detection results. In this scenario, the inspection report obtained by executing steps S110-S130 based on the abnormal item information can be used as an updated inspection report. It should be noted that the judgment conditions of the abnormal detection results can be customized according to actual needs, and the embodiments of the present application do not make specific limitations.

[0031] In some embodiments, the step of retrieving documents semantically related to the abnormal item information from a predetermined knowledge base as the context information of the abnormal item information in the above step S110 may specifically include: performing data cleaning processing on the abnormal item information to obtain cleaned abnormal item information; performing vectorization processing on the cleaned abnormal item information to obtain a text semantic vector; searching the knowledge base based on the cleaned abnormal item information and the text semantic vector respectively to obtain a first search result and a second search result; generating the context information of the abnormal item information based on the first search result and the second search result.

[0032] Exemplarily, data cleaning processing is used to remove irrelevant information and noise data in abnormal item information. Data cleaning processing may include at least one of the following processing: removing stop words and removing special characters. Among them, removing stop words means: removing words that do not represent specific things or concepts, such as function words (conjunctions, prepositions, pronouns, etc.), emoticons, modal particles, etc. Removing special characters means: removing symbols such as punctuation marks, currency symbols, and mathematical symbols.

[0033] Exemplarily, the cleaned abnormal item information can also be segmented to obtain multiple words in the abnormal item information. The word segmentation process refers to: dividing the continuous text string in the cleaned abnormal item information into separate words or phrases. The word segmentation can be segmented according to spaces, punctuation marks, etc., or the word segmentation model can be used for word segmentation. The specific method of word segmentation can be selected according to actual needs, and the embodiments of the present application do not make specific limitations.

[0034] Exemplarily, the first search result is obtained by searching the knowledge base based on the cleaned abnormal item information. In this example, the cleaned abnormal item information can be directly matched with the document segments of each technical document in the knowledge base, and the document segment containing the abnormal item information is used as the first search result. Alternatively, after the cleaned abnormal item information is segmented, the multiple words of the abnormal item information obtained by the segmentation are matched with the document segments of each technical document in the knowledge base, and the document segment containing at least one of the words is used as the first search result. Alternatively, keywords can be extracted from the cleaned abnormal item information first, and the extracted keywords can be matched with the document segments of each technical document in the knowledge base, and the document segment containing the keyword can be used as the first search result.

[0035] Exemplarily, the technical documents in the knowledge base can be structured or semi-structured documents. That is to say, the technical documents can organize information according to a certain logic and format to make them have structured characteristics. Semi-structured is a form of structuring. Semi-structured technical documents have certain structural information, but this information is not as fixed and strict as structured data, and has certain flexibility and scalability. Exemplarily, semi-structured documents can have at least one of the following file types: Extensible Markup Language (XML), Lightweight Data Interchange Format (JavaScript Object Notation, JSON), Hypertext Markup Language (Hyper Text Markup Language, HTML) type documents, etc.

[0036] For example, for structured or semi-structured technical documents in the knowledge base, the embodiment of the present application can provide a semantic slicing method, which can slice the technical documents into document fragments while ensuring semantic integrity. In other words, the technical documents can be sliced ​​according to the document title, paragraph title, text, table, and picture to obtain the document fragments of each technical document.

[0037] In this example, the keyword is the carrier of key information in the abnormal item information, and can be used to indicate the text features of the abnormal item information. Using keywords to search in the knowledge base helps to retrieve the document with the most matching text features from the knowledge base. Among them, the text features include but are not limited to at least one of the following information: named entity type, part of speech, and word frequency. The named entity type is used to identify entities with specific meanings from the abnormal item information, for example, including but not limited to at least one of the following: time, date, abnormal item name, and proper name. Part of speech refers to the grammatical role of a word in a sentence, and the part of speech may include at least one of the following: noun, verb, adjective, adverb, pronoun, etc. Word frequency refers to the number of times a word appears in the corpus data. Based on this, the keyword can be at least one of the following in the abnormal item information: a predetermined entity, a word with a predetermined part of speech, and a word with a word frequency greater than or equal to a predetermined word frequency threshold.

[0038] For example, the abnormal item information can be preprocessed and keywords can be extracted based on the natural language processing toolkit (Han Language Processing, HanLP). The natural language processing toolkit can be used to perform various natural language processing tasks, including word segmentation, stop word removal, special character removal, part-of-speech tagging, named entity recognition and other language processing functions.

[0039] Exemplarily, the cleaned abnormal item information is vectorized to obtain a text semantic vector, which may specifically include: directly vectorizing the cleaned abnormal item information to obtain a text semantic vector. Alternatively, firstly segment the cleaned abnormal item information to obtain multiple words; then, vectorize the multiple words to obtain multiple word vectors; then, generate a text semantic vector based on the multiple word vectors. Specifically, generating a text semantic vector based on multiple word vectors may include multiple ways. For example: multiple word vectors may be vector averaged to obtain a text semantic vector. For another example: multiple word vectors may be weighted averaged to obtain a text semantic vector, and the weight of any word vector is proportional to the word frequency of the word corresponding to the word vector, and the word frequency is used to represent: the number of times the corresponding word appears in the abnormal item information.

[0040] Exemplarily, the knowledge base is searched based on the text semantic vector, and the second search result is obtained. In this example, the text semantic vector can be used to search for the semantic vectors of each technical document in the knowledge base, and the documents to which the semantic vectors with vector similarity greater than a predetermined first similarity threshold belong are obtained, and the document fragments of the obtained documents are used as the second search result. Alternatively, the text semantic vector can be used to search for the semantic vectors of the document fragments of each technical document in the knowledge base, and the document fragments to which the retrieved semantic vectors with vector similarity greater than a predetermined second similarity threshold belong are used as the second search result.

[0041] The predetermined first similarity threshold and the predetermined second similarity threshold may be the same or different, and may be customized according to actual needs, and are not specifically limited in the embodiments of the present application.

[0042] In this example, the former retrieval method requires converting all technical documents into document-level semantic vectors as a whole, which is suitable for application scenarios with relatively short documents (for example, the number of words is less than or equal to a predetermined word count threshold and / or the number of pages is less than or equal to a page count threshold) to improve retrieval efficiency. The latter retrieval method requires dividing the technical documents into smaller document fragments, which are then vectorized to obtain document fragment-level semantic vectors, which is suitable for scenarios with longer documents (for example, the number of words is greater than a predetermined word count threshold and / or the number of pages is greater than a predetermined page count threshold) to provide more fine-grained matching and improve retrieval accuracy.

[0043] Exemplarily, there are multiple implementation methods for generating context information of the abnormal item information based on the first search result and the second search result. For example: merging the document fragments in the two into a document fragment set, then deduplicating, and selecting at least one document fragment from the deduplicated document fragments as context information; or merging the document fragments in the two into a document fragment set, and then sorting the document fragments in the document fragment set. The sorting rule may be, for example, sorting based on the text similarity between each document fragment and the cleaned abnormal item information (either ascending or descending order), and selecting the first at least one document fragment from the sorted document fragments in descending order of text similarity as context information.

[0044] In the description of the above embodiments, retrieval based on the cleaned abnormal item information is also called text retrieval, sparse retrieval or retrieval based on text matching; retrieval based on text semantic vectors is also called vector retrieval, dense retrieval or semantic retrieval.

[0045] In an embodiment of the present application, by performing data cleaning processing on the abnormal item information, irrelevant information or redundant information in the abnormal item information can be removed, thereby improving data processing efficiency. The abnormal item information is divided into multiple words through word segmentation, so that the abnormal item information can be divided into smaller processing units, and then the multiple words obtained by the division are vectorized to obtain multiple word vectors, and then a text semantic vector is generated based on the multiple word vectors, which is conducive to reducing the complexity of text processing and improving processing efficiency. In addition, based on the abnormal item information after cleaning, the knowledge base can be accurately matched to the query results, and the knowledge base can be searched separately based on the text semantic vector, which can provide more relevant query results by capturing the deep semantic features of the text. Combining the retrieval results of the two can improve the accuracy of the retrieval results and provide richer and more accurate query results.

[0046] In some embodiments, the knowledge base includes at least one document segment of a technical document and a semantic vector of the corresponding document segment, and each technical document includes at least one of a historical inspection report and a product document; the document segment of each technical document carries dependency information, and the dependency information is used to indicate the hierarchical structure of the corresponding document segment in the corresponding technical document. The above-mentioned steps of searching the knowledge base based on the preprocessed abnormal item information and the text semantic vector to obtain the first search result and the second search result may specifically include the following steps: performing keyword matching in the document segments of each technical document based on the keywords in the cleaned abnormal item information, and taking the matched document segment as the first search result; performing vector search in the semantic vector of the corresponding document segment using the text semantic vector, and taking the document segment corresponding to the retrieved similar semantic vector as the second search result, and the similarity between the similar semantic vector and the text semantic vector is greater than or equal to the predetermined similarity threshold.

[0047] Exemplarily, the method of extracting keywords and performing keyword matching in the document allocation of each technical document based on the keywords can be found in the description of the above embodiment, which will not be described in detail here.

[0048] Exemplarily, the dependency information is used to indicate the hierarchical structure of the document fragments in the technical document to which it belongs. Exemplarily, if a technical document includes a document title, at least one paragraph title, text, tables and pictures under each paragraph title, then the dependency information of the document fragments in the technical document includes at least one of the following: parent fragment information, child fragment information and sibling fragment information.

[0049] Specifically, the parent fragment information refers to: the title of the previous level of the current document fragment, such as at least one of the document title and paragraph title. The child fragment information refers to: the content of the next level of the current document fragment, such as subtitles, text paragraphs, table fields and pictures. The sibling fragment information refers to: other document fragments at the same level as the current document fragment, that is, other child fragment information under the same parent fragment information. Through these dependency information, the hierarchical structure and mutual relationship between document fragments can be constructed and indicated.

[0050] In this example, the document fragments of each technical document may carry dependency information of the corresponding document fragments. The semantic vectors of the document fragments of each technical document may be the semantic vectors of the corresponding document fragments carrying the dependency information. Adding the corresponding dependency information to the document fragments is conducive to maintaining the semantic integrity of the document fragments, thereby improving the retrieval accuracy and recall rate when performing keyword matching and vector retrieval.

[0051] Exemplarily, a semantic vector model (Embedding Model) can be used to vectorize the cleaned abnormal item information to obtain a text semantic vector. The semantic vector model is used to map text, words or other high-dimensional data into a continuous vector space to capture its semantic information. The semantic vector model can be at least one of the following: a general semantic vector model (BGE-M3), a bidirectional encoder representation from transformers (Bidirectional Encoder Representations from Transformers, BERT) model, a bilingual and cross-lingual embedding model library (Bilingual and Crosslingual Embedding, BCE), etc. The semantic vector model can be selected according to actual needs, and the embodiments of the present application do not make specific limitations.

[0052] Exemplarily, if the technical documents in the knowledge base are structured or semi-structured documents, the semantic vectors of the document segments of the technical documents can be obtained by the following steps. First, the technical documents can be segmented according to the document titles, paragraph titles, texts, tables, and pictures to obtain the document segments of the technical documents; secondly, each document segment obtained by segmentation is input into a semantic vector model, such as a general semantic vector model, to generate semantic vectors for each document segment. In the above steps, if the length of the document segment input into the semantic vector model exceeds the maximum input length of the semantic vector model, the input document segment can be truncated first, and then the truncated document segment can be vectorized to obtain the semantic vector of the corresponding document segment.

[0053] In this embodiment, based on the keywords of the cleaned abnormal item information, a first retrieval result is obtained by keyword matching retrieval in the document segments of each technical document in the knowledge base, and based on the text semantic vector of the cleaned abnormal item information, a second retrieval result is obtained based on vector similarity retrieval from the semantic vectors of the document segments of each technical document in the knowledge base, thereby providing a data basis for subsequently generating context information based on the first retrieval result and the second retrieval result.

[0054] In an embodiment of the present application, for abnormal item information in the inspection report (or abnormal item information after cleaning), the abnormal item information is input into the semantic vector model for vectorization processing to obtain a text semantic vector, and the abnormal item information and the text semantic vector are used to search the knowledge base, and the context information of the abnormal item information is generated based on the retrieval results of the two, which is conducive to improving the accuracy of the retrieval results.

[0055] In some embodiments, the step of generating context information of the abnormal item information according to the first search result and the second search result can be implemented by weighted fusion screening of the two search results in addition to the two implementation methods exemplified in the above embodiments. The specific implementation method is described in detail below.

[0056] In some embodiments, the step of generating context information of abnormal item information based on the first retrieval result and the second retrieval result may specifically include: determining each first weighted similarity based on each first similarity and a preset first weight, each first similarity being: the semantic similarity between each document fragment in the first retrieval result and the keyword; determining each second weighted similarity based on each second similarity and a preset second weight, each second similarity being: the vector similarity between the semantic vector of each document fragment in the second retrieval result and the text semantic vector; selecting at least one document fragment with the largest weighted similarity from the first retrieval result and the second retrieval result to generate context information.

[0057] Exemplarily, the first weight refers to: a weight pre-assigned to each document fragment in the first search result, and the second weight refers to: a weight pre-assigned to each document fragment in the second search result. Each first similarity is weighted using the first weight to obtain each first weighted similarity, and each second similarity is weighted using the second weight to obtain each second weighted similarity; each document fragment in the first search result and each document fragment in the second search result are sorted according to their corresponding weighted similarities (in ascending or descending order) to obtain a sorting result. Among them, the first weight and the second weight can be customized according to actual conditions, and the embodiments of the present application do not make specific limitations.

[0058] In this embodiment, after obtaining the sorting result, at least one document fragment with the largest weighted similarity can be selected from the sorting result as context information. Specifically, for each document fragment in the first search result and each document fragment in the second search result, if they are arranged in descending order according to their corresponding weighted similarities, at least one document fragment can be selected in sequence from the first document fragment in the sorting result in the direction of decreasing similarity as context information. If they are arranged in ascending order according to their corresponding weighted similarities, at least one document fragment can be selected in sequence from the last document fragment in the sorting result in the direction of increasing similarity as context information. In this embodiment, selecting at least one most similar document fragment from the sorting result as context information is conducive to extracting context information that is most relevant to the semantics of the abnormal item information.

[0059] In some embodiments, the above-mentioned step S120 may specifically include: merging each document fragment in the context information with the corresponding dependency information to obtain relevant documents of the context information; adding the relevant documents of the context information to the abnormal item information to obtain merged abnormal item information; based on the merged abnormal item information, using a large language model to perform abnormal summary on the inspection items to obtain abnormal item summary text.

[0060] Exemplarily, through the description of dependency information in the above embodiment, it can be known that the dependency information of the document fragments in the technical document includes at least one of the following: parent fragment information, child fragment information, and sibling fragment information. Therefore, the merging means: merging each document fragment in the context information with at least one of the following information: parent fragment information, at least one child fragment information, and at least one sibling fragment information of the corresponding document fragment.

[0061] In this example, Large Language Models (LLM) are referred to as large models or large language models. Large language models have certain context understanding capabilities and the ability to identify, reason and summarize abnormal item information. Large models can be used to identify and summarize abnormal items based on the merged abnormal item information and generate high-quality abnormal item summary text.

[0062] The large model of one or more embodiments of the present application can be a large language model based on deep learning, which contains more than one billion parameters, and the underlying converter of the large model contains a series of neural networks. In one or more embodiments, an encoder and / or a decoder may be included, and have a self-attention function. These two types of modules extract meaning from the input text and clarify the structural relationship of context words. The model is pre-trained on a public dataset containing more than 1 terabyte (TB) of text data to learn general language representation.

[0063] In this embodiment, the merged abnormal item information includes: abnormal item information, context information, and dependency information corresponding to each document segment in the context information. It can be seen that the merged abnormal item information is abnormal item information containing rich context information, and the dependency information is conducive to ensuring the information integrity of the abnormal item information. Using a large language model to summarize the abnormalities based on the merged abnormal item information is conducive to obtaining a clear and easy-to-understand abnormal item summary text, reducing the cost of understanding the inspection report for operation and maintenance personnel, and improving the effectiveness and readability of the abnormal item information in the inspection report.

[0064] In an embodiment of the present application, based on querying context information from a knowledge base and summarizing exceptions based on a large language model, the method of this embodiment provides a method that combines information retrieval and generation models (Retriever-Augmented Generation, RAG), and utilizes the context information most relevant to the semantics of the abnormal item information queried from the knowledge base, and the context dependency information obtained from the knowledge base to enhance the amount of information and the semantic understanding of the abnormal item information in the inspection report, and then utilizes the large model to summarize the exceptions to obtain a clear and easy-to-understand summary text of the abnormal item, thereby providing a data basis for subsequently reducing the difficulty of interpreting the inspection report, making the report easier to understand, and thereby improving the readability of the report.

[0065] In some embodiments, the large language model is a language model that is fine-tuned based on a pre-constructed standard data set, and the standard data set includes: various abnormal scenarios of the detection object and standardized description information of each abnormal scenario; the above-mentioned step of using the large language model to perform abnormal summary of the inspection items based on the merged abnormal item information to obtain the abnormal item summary information can specifically include: using the fine-tuned language model to generate target description information based on the merged abnormal item information, the target description information is: standardized description information of the abnormal scenario corresponding to the merged abnormal item information; generating abnormal item summary information based on the target description information.

[0066] In order to adapt to the specific application scenario of the present invention - the cloud network networking scenario, we fine-tuned the model on a dedicated data set (i.e., the standard data set in this embodiment) containing thousands of articles for standardized descriptions of abnormal scenarios in the inspection field in the cloud network networking scenario. This model receives the merged abnormal item information as input and can generate concise and accurate standardized description information of abnormal scenarios as output. Compared with the simple listing of abnormal items in traditional inspection results, the advantage of using a large language model in the present invention is that it can better capture the complex relationships in the text and generate more natural and fluent standardized description information of abnormal scenarios.

[0067] Exemplarily, a standard data set may be a data set in the field of cloud network inspection, which contains standardized description information of abnormal scenarios of cloud products such as load balancing and cloud enterprise network. The data set may be constructed in advance by experts, including at least one abnormal scenario of each of the above cloud products and corresponding description information of each abnormal scenario. Each abnormal scenario may include at least one of the following scenarios: an abnormality occurs when the server attempts to process a request, network address security protocol negotiation fails, etc. Among them, the abnormality that occurs when the server attempts to process a request can be collectively referred to as a "5xx error", and "5xx" is the beginning number of a series of HyperText Transfer Protocol (HTTP) status codes. For example, a general server error (500 Internal Server Error), a server currently unavailable error (503 ServiceUnavailable), etc. Each data entry in the data set is verified by domain experts to ensure the high quality and accuracy of each data entry.

[0068] In an embodiment of the present application, fine-tuning means that on the basis of a pre-trained language model, the model parameters are further adjusted using a standard data set in the field of cloud network inspection, so that the model is more suitable for processing tasks in the field of cloud network inspection, thereby reducing the model hallucination phenomenon, that is, reducing the probability of the model outputting erroneous information, and obtaining a model suitable for the vertical field of cloud network inspection report summary, that is, a language model dedicated to the field of cloud network inspection report.

[0069] Exemplarily, the fine-tuning model training process involves adjusting the weights and / or model parameters of the language model. After the model fine-tuning is completed, the merged abnormal item information is input into the fine-tuned language model for summary to obtain standardized description information of the abnormal scenario corresponding to the merged abnormal item information, and the standardized description information obtained by the summary can be used as the abnormal item summary information.

[0070] For example, the existing language model can be fine-tuned using Low-Rank Adaptation (LoRA) technology. Low-rank adaptation technology is a lightweight model fine-tuning technology. Its principle is to adjust the model parameters by introducing a low-rank matrix. This method can reduce the number of parameters that need to be updated while maintaining the model performance, which is conducive to reducing computational complexity, saving computing resources and improving training efficiency, making the model fine-tuning process more efficient.

[0071] It should be understood that other model fine-tuning techniques can also be used to fine-tune the language model, such as full model fine-tuning (Full Fine-tuning), partial fine-tuning (Partial Fine-tuning), etc. Full model fine-tuning can update all parameters of the current existing model based on the standard data set. Its characteristic is that the model can fully learn the relevant knowledge based on the standard data set, so as to adapt to the inspection item anomaly summary task based on the standard data set and improve its performance on this task. Partial fine-tuning can only update some parameters of the model based on the standard data set. Its characteristic is that the general knowledge of the current existing model can be retained. In this technology, fine-tuning is based on the standard data set to adapt to specific tasks, such as the inspection item anomaly summary task based on the standard data set in the embodiment of the present application. Therefore, in actual application scenarios, the model fine-tuning technology can be selected according to actual needs, and the embodiment of the present application does not make specific limitations.

[0072] In this embodiment, the current existing language model is fine-tuned based on the standard data set to obtain a fine-tuned language model, thereby improving the model's ability to follow instructions for the inspection item exception summary task. The fine-tuned language model can be used to perform exception summaries on the inspection items based on the merged exception item information. The exception item summary information can further improve the understanding of the exception item information in the inspection report, and can improve the logical coherence and information richness of the inspection report, thereby reducing the workload of operation and maintenance personnel and improving the overall operation and maintenance efficiency.

[0073] In some embodiments, the above-mentioned step of generating abnormal item summary information based on the target description information may specifically include: obtaining preset prompt words, the prompt words are used to guide the fine-tuned language model to perform cluster analysis of abnormal items of the same type on abnormal scenes according to the target description information; using the fine-tuned language model, based on the prompt words and the target description information, performing cluster analysis of abnormal items of the same type on the abnormal scenes to which the target description information belongs, to obtain standardized description information grouped according to abnormal items of the same type; generating an abnormal item summary text according to the standardized description information grouped according to abnormal items of the same type.

[0074] Exemplarily, a prompt is a sample text used to guide the fine-tuned language model to perform a specific task. In the embodiment of the present application, the target description information is the standardized description information of the abnormal scene corresponding to the merged abnormal item information. Based on the prompt, the fine-tuned language model can be guided by specific examples to identify the commonalities of abnormal items of the same type from the standardized description information of the abnormal scene, and cluster the abnormal items accordingly.

[0075] In this embodiment, by fine-tuning the language model, abnormal items of the same type are clustered together based on prompt words and target description information, which can further improve the induction and summary of abnormal item information in the inspection report, facilitate the subsequent rapid location of the root cause of the abnormality based on abnormal items of the same type, and improve the diagnosis and processing efficiency of abnormal problems.

[0076] In some embodiments, the knowledge base includes a problem solution corresponding to standardized description information of at least one abnormal scenario; after the above step S130, the method also includes: retrieving a problem solution corresponding to the target description information from the knowledge base; and adding the retrieved problem solution to the inspection report.

[0077] For example, the problem solutions in the knowledge base are problem solutions that have been established and verified by domain experts in the past. The target description information is input into the knowledge base for retrieval to obtain the corresponding problem solutions. The retrieved problem solutions are added to the inspection report. The inspection report finally generated contains not only a detailed summary of abnormal items, but also specific solutions to the problems, so that technicians and operation and maintenance personnel can not only find abnormal problems during the inspection process, but also solve them quickly. Through this process, not only can the effectiveness and practicality of the inspection report be improved, but it can also greatly reduce the workload of operation and maintenance personnel and improve the overall operation and maintenance efficiency.

[0078] In this embodiment, for the abnormal item information in the inspection report, the context information retrieved from the knowledge base and the dependency information of the document segment obtained from the knowledge base can be used to enrich the abnormal item information, and the abnormal items can be summarized using the fine-tuned large language model. The final inspection report includes not only the summarized abnormal item information, but also recommended solutions, which is helpful to assist users to quickly locate and solve problems.

[0079] In some embodiments, before step S110, the method also includes: obtaining at least one technical document and a problem solution of the detection object, each technical document including at least one of a historical inspection report and a product document; segmenting each technical document based on the hierarchical structure of each technical document to obtain document segments of each technical document; adding dependency information to the document segments of each technical document, the dependency information is used to indicate the hierarchical structure of the corresponding document segment in the corresponding technical document; vectorizing the document segments with the dependency information added to obtain semantic vectors of the corresponding document segments; generating a knowledge base based on the document segments with the dependency information added, the semantic vectors of the corresponding document segments and the problem solution.

[0080] Exemplarily, in order to build a knowledge base, technical documents and problem solutions can be collected in advance. Technical documents can be segmented based on their hierarchical structure to obtain corresponding document segments, and their dependency information is carried in the corresponding document segments. Then, the document segments carrying the dependency information are vectorized to obtain semantic vectors of the corresponding document segments. Then, a knowledge base is generated based on the document segments carrying dependency information of each technical document, the corresponding document segment semantic vectors, and the problem solution. It can be seen from the description of the above embodiments that adding corresponding dependency information to the document segments is conducive to maintaining the semantic integrity of the document segments, thereby improving the retrieval accuracy and recall rate when performing keyword matching and vector retrieval.

[0081] Exemplarily, the above-mentioned document fragments, semantic vectors corresponding to the document fragments, and problem solutions contained in the knowledge base can be stored in a distributed search and analysis engine. Exemplarily, the distributed search and analysis engine can be used for information storage and retrieval. For information retrieval, the distributed search and analysis engine supports text retrieval and vector retrieval.

[0082] As a specific example, the distributed search and analysis engine stores the document segments of each technical document and the semantic vectors of the corresponding document segments. For text retrieval, the distributed search and analysis engine can perform keyword matching in the document segments of each technical document based on the keywords in the cleaned abnormal item information; for vector retrieval, the text semantic vector of the abnormal item information can be used to perform vector retrieval from the semantic vector in the corresponding document segment through the distributed search and analysis engine.

[0083] Exemplarily, the distributed search and analysis engine can be a full-text search engine (Elasticsearch), which is an open source, highly scalable search and analysis engine for fast, real-time full-text retrieval and data analysis. Exemplarily, the distributed search and analysis engine can also be any of: a machine data engine (splunk) and a distributed near-real-time searchable database (Easysearch). In actual application scenarios, the type of distributed search and analysis engine can be selected according to actual needs, and the embodiments of this application do not make specific limitations.

[0084] In this embodiment, the knowledge base includes not only the document segments and problem solutions of the technical documents, but also the semantic vectors of the document segments, which provide a data basis for efficiently and accurately retrieving information from the knowledge base in the subsequent steps. In addition, the document segments of each technical document carry dependency information, which is conducive to maintaining the semantic integrity of the document segments, thereby improving the retrieval accuracy and recall rate when performing keyword matching and vector retrieval.

[0085] In some embodiments, before the step of adding dependency information to the document slices of each technical document, the method further includes: generating dependency information of the document slices of each technical document according to the hierarchical structure of the document slices of each technical document in the corresponding technical document.

[0086] In this embodiment, the dependency information between document fragments can be saved in the knowledge base. The dependency information is established based on the hierarchical structure of the document fragments of each technical document in the technical document to which it belongs. By understanding the hierarchical structure of the document fragments in the technical document, it is helpful to enhance the information integrity of the abnormal item information, and it can also more accurately capture the semantic association between the document fragments, thereby helping to reduce the probability of inaccurate retrieval results due to incomplete information, and improve the retrieval efficiency and the accuracy of the retrieval results.

[0087] According to the inspection report generation method for the cloud network networking scenario of the embodiment of the present application, one or more products of the cloud network networking are taken as the detection objects. For the abnormal item information of the detection object, the semantically related documents retrieved from the knowledge base can be used as the context information of the abnormal item information, and the abnormal item information and its context information are abnormally summarized using the large language model, and the inspection report is generated using the obtained abnormal item summary text. In this method, by using the documents semantically related to the abnormal item information retrieved from the knowledge base as the context information, the semantic understanding of the abnormal item information by the large language model can be better enhanced. On the basis of enhancing the semantic understanding, the abnormal item summary text generated by the large language model helps to more accurately describe the abnormal item information, which is conducive to improving the accuracy and readability of the inspection report, reducing the difficulty of interpreting the inspection report, making the content of the inspection report easier to understand, and thus improving the quality of the inspection report.

[0088] Figure 2 A flow chart of a method for generating an inspection report for a cloud network provided in an embodiment of the present application is shown. Figure 2 As shown, the method may include the following steps S210-S250.

[0089] S210, obtaining an abnormality detection result of any inspection item from the inspection results of the cloud network networking product.

[0090] S220, generating abnormal item information according to the abnormality detection result.

[0091] S230, based on the abnormal item information, retrieve documents semantically related to the abnormal item information from a predetermined knowledge base as context information of the abnormal item information.

[0092] S240, based on the context information and the abnormal item information, using a predetermined large language model to perform abnormal summary on the inspection items to obtain an abnormal item summary text.

[0093] S250, generating an inspection report of the cloud network networking product by using the abnormal item summary text.

[0094] According to the method of the embodiment of the present application, for the abnormal detection results of each inspection item in the inspection results of the cloud network networking product, abnormal item information can be generated based on the abnormal detection results, and documents related to the abnormal item information can be retrieved from the knowledge base as context information of the abnormal item information, and the abnormal item information and its context information are abnormally summarized using a large language model, and an inspection report is generated using the summarized abnormal item summary text. In this method, by using documents semantically related to the abnormal item information retrieved from the knowledge base as context information, the semantic understanding of the abnormal item information by the large language model can be better enhanced. On the basis of enhancing the semantic understanding, the abnormal item summary text generated by the large language model helps to more accurately describe the abnormal item information, which is beneficial to improving the accuracy and readability of the inspection report, reducing the difficulty of interpreting the inspection report, making the content of the inspection report easier to understand, and thus improving the quality of the inspection report.

[0095] Exemplarily, the cloud network networking product is one or more products of the cloud network networking. The specific types of products can refer to the description of the above embodiment, which will not be repeated here.

[0096] In some embodiments, the inspection report includes standardized description information of abnormal scenarios corresponding to the abnormal item information; after the above step S250, the method also includes: retrieving problem solutions corresponding to the standardized description information from the knowledge base; and adding the retrieved problem solutions to the inspection report.

[0097] In this embodiment, the retrieved problem solutions are added to the inspection report, and the inspection report finally generated contains not only a detailed summary of abnormal items, but also specific solutions to the problems, so that technicians and operation and maintenance personnel can not only find abnormal problems during the inspection process, but also solve the problems quickly. Through this process, not only can the effectiveness and practicality of the inspection report be improved, but it can also help greatly reduce the workload of operation and maintenance personnel and improve the overall operation and maintenance efficiency.

[0098] According to the method of the embodiment of the present application, by summarizing the abnormal items in the abnormal detection results of the cloud network networking products and providing solutions, an intelligent operation and maintenance solution is provided for the complex networking of the cloud network, which is conducive to helping operation and maintenance personnel quickly interpret inspection reports, find the root cause of the abnormality, and improve the automation and intelligence capabilities of cloud network operation and maintenance based on the provided solution.

[0099] Figure 3 A flow chart showing a method for generating an inspection report for a cloud network networking according to an exemplary embodiment of the present application is shown. Figure 3 As shown, in some embodiments, the method may include the following steps.

[0100] S310, knowledge base construction.

[0101] Specifically, step S310 may include the following steps.

[0102] S311, information collection.

[0103] In this step, in order to build a knowledge base, it is necessary to collect and organize technical documents and solutions in the field of cloud network networking in advance. The technical documents include at least one of historical inspection reports and product documents. The technical documents and solutions can each be a structured or semi-structured document.

[0104] S312, semantic segmentation.

[0105] In this step, under the premise of ensuring semantic integrity, the technical documents in the knowledge base are cut into document fragments of a length suitable for input into the semantic vector model. Exemplarily, the technical documents can be sliced ​​according to the document title, paragraph title, text, table, and picture of the technical document, and the dependency information between the document fragments is maintained, such as the document title and the paragraph title are parent-child relationships, the document title is the parent fragment information, and the paragraph title is the child fragment information. The parent fragment information is added to the child fragment information to improve the subsequent retrieval recall rate and accuracy.

[0106] S313, vectorization processing.

[0107] In this step, each document fragment can be vectorized to generate a semantic vector. If the length of the input document fragment exceeds the maximum input length of the semantic vector model, the input document fragment is truncated. The input document fragment can carry corresponding dependency information.

[0108] S314, storing the document slices and their semantic vectors.

[0109] In this step, the document fragments carrying dependency information and their corresponding semantic vectors can be stored in a database distributed search and analysis engine that supports text retrieval (also known as sparse retrieval) and vector retrieval (also known as dense retrieval) to complete the construction of the knowledge base. The distributed search and analysis engine is conducive to the subsequent efficient and accurate retrieval of information from the knowledge base.

[0110] S320, inspection report processing.

[0111] Specifically, step S320 may include the following steps.

[0112] S321, data cleaning processing.

[0113] In this step, for any abnormal item information in the inspection report, the natural language processing toolkit can be used to perform data cleaning on the abnormal item information to obtain the cleaned abnormal item information. Then, the semantic vector model can be used to vectorize the cleaned abnormal item information to generate a text semantic vector of the abnormal item information.

[0114] S322, query enhancement.

[0115] In this step, query enhancement is then performed on the abnormal item information. Query enhancement is equivalent to reducing ambiguity and expressing query intent more accurately through rewriting. Exemplarily, assuming that the abnormal item information is bandwidth package excess packet loss, it can be rewritten through query enhancement as, for example: bandwidth package excess packet loss solution. Exemplarily, the currently existing large language model can be used to perform query enhancement on the abnormal item information to obtain enhanced abnormal item information for use in constructing a more accurate query expression based on the abnormal item information. It should be understood that the type of large language model for query enhancement can be selected according to actual needs, and the embodiments of the present application do not make specific limitations.

[0116] S323, hybrid search.

[0117] In this step, data cleaning is first performed on the enhanced abnormal item information. The steps of data cleaning refer to the specific steps of data cleaning for the abnormal item information described in the above embodiment, which will not be repeated here.

[0118] After data cleaning of the enhanced abnormal item information, the cleaned abnormal item information and the semantic vector of the abnormal item information are used for text retrieval and vector retrieval. As an example, the cleaned abnormal item information can be used to perform keyword matching on the document fragments of each technical document through the inverted index mechanism of the distributed search and analysis engine, and a dense retrieval algorithm based on the vector space model (VSM) can be used to search for semantically similar documents in descending order of similarity from the semantic vectors of the document fragments of each technical document, and the retrieval results of the two are weighted and sorted to obtain the final retrieval results.

[0119] In this example, the inverted index mechanism is an indexing method used to establish a mapping relationship between first information and second information, where the first information is any word appearing in a document segment, and the second information is all document segments where the word appears. The efficiency of text queries can be effectively improved through inverted indexes. In this example, the dense retrieval algorithm can be an Approximate Nearest Neighbor (ANN) algorithm, which can find data points (i.e., the found semantic vectors) that are very close to the query point (i.e., the semantic vectors of the document segments of each technical document) in the data set (i.e., the semantic vectors of the document segments of each technical document), thereby minimizing the amount of calculation in the search process and improving retrieval efficiency while ensuring a certain degree of accuracy.

[0120] Next, the document fragments in the final retrieval results can be parsed and screened. For example, by re-ranking, at least one document fragment can be selected in descending order of text similarity as the context information of the abnormal item information. The context information of each selected document fragment is then merged with the dependency information of the corresponding document fragment obtained when the knowledge base is constructed to obtain merged related documents. The merged related documents are integrated into the abnormal item information, which is conducive to enhancing the information volume and situational background of the abnormal item information. Subsequently, the abnormal item information can be reorganized and optimized based on the merged related documents to ensure its logical coherence, emotional consistency and compliance with inspection report specifications, thereby obtaining abnormal item text containing rich context information.

[0121] S330, summary of abnormal items and recommended solutions.

[0122] Specifically, step S330 may include the following steps.

[0123] S331, obtain the abnormal item text.

[0124] In this step, the abnormal item text containing rich context information obtained after the above-mentioned hybrid retrieval is obtained. The abnormal item text is a text carrying context information and corresponding dependency information, hereinafter referred to as enriched abnormal item text or merged abnormal item text.

[0125] S332, model summary.

[0126] In this step, the enriched abnormal item text is input into the large language model, which is a language model fine-tuned using a pre-built standard data set. The model is used to summarize the abnormal items based on the merged abnormal item text to obtain the abnormal item summary text. The abnormal item summary text is generated based on the target description information, which includes: standardized description information of the abnormal scenario corresponding to the merged abnormal item text.

[0127] S333, search solution.

[0128] In this step, the problem solution corresponding to the target description information can be retrieved from the knowledge base.

[0129] S334, generating a final inspection report.

[0130] In this step, the summary text of the abnormal items and the retrieved problem solutions are added to the inspection report to generate the final inspection report.

[0131] In the method of the embodiment of the present application, a knowledge base including historical inspection reports, product documents and solutions is constructed, and semantic slicing and vector processing technology are applied to perform efficient data storage and retrieval in the knowledge base. Based on the standard data set, the current existing language model is fine-tuned to obtain a fine-tuned language model, and the model's ability to follow instructions for the abnormal summary task of the inspection items of each product of the cloud network networking is improved. The fine-tuned language model is used to perform abnormal summaries of the inspection items based on the merged abnormal item information to generate a clear and easy-to-understand summary text. Subsequently, a specific solution for abnormal item information is provided in combination with pre-built solutions and knowledge base retrieval technology. The method of the embodiment of the present application is conducive to significantly improving the effectiveness and readability of cloud network networking inspection reports, reducing the understanding cost of operation and maintenance personnel, and providing a complete closed loop from problem discovery to problem solving, which greatly reduces the burden of operation and maintenance work, improves the overall operation and maintenance efficiency, and realizes the intelligent and efficient management of cloud network networking.

[0132] In the embodiment of the present application, a knowledge base is constructed based on historical inspection reports, product documents, and solutions of the cloud network. According to the large amount of abnormal item information in the inspection report, the context information retrieved from the knowledge base and the dependency information of the document fragments maintained in the knowledge base are used to enrich it, and the abnormal items are summarized using a fine-tuned large language model, and solutions are recommended for the summarized abnormal items, providing an intelligent and efficient solution for the inspection of complex cloud network networking, and promoting the intelligent development of cloud computing. The method of the embodiment of the present application combines natural language processing technology and large language model technology, applies these technologies to the inspection report summary and problem solving field of cloud network networking, provides automated and intelligent solutions with high feasibility, realizes the summary of cloud network networking inspection reports and the recommendation of targeted solutions, is conducive to quickly obtaining key anomalies and corresponding solutions in cloud network networking, and ensures the stability of cloud network networking.

[0133] Corresponding to the application scenario and method of the inspection report generation method for the cloud network networking scenario provided in the embodiment of the present application, the embodiment of the present application also provides an inspection report generation device for the cloud network networking scenario.

[0134] Figure 4 The structure diagram of the inspection report generation device for the cloud network networking scenario provided by the embodiment of the present application is shown. Figure 4 As shown, the device includes the following modules.

[0135] The retrieval module 410 is used to retrieve documents semantically related to the abnormal item information from a predetermined knowledge base based on the acquired abnormal item information as the context information of the abnormal item information; the abnormal item information is used to indicate the abnormal detection result of any inspection item of the detection object, and the detection object is one or more products used for cloud network networking; the summary module 420 is used to perform abnormal summary of the inspection items based on the context information and the abnormal item information using a predetermined large language model to obtain the abnormal item summary text; the generation module 430 is used to generate an inspection report using the abnormal item summary text.

[0136] In some embodiments, when the retrieval module 410 is used to retrieve documents semantically related to the abnormal item information from a predetermined knowledge base as context information of the abnormal item information, it is specifically used to: perform data cleaning processing on the abnormal item information to obtain cleaned abnormal item information; perform vectorization processing on the cleaned abnormal item information to obtain a text semantic vector; search the knowledge base based on the cleaned abnormal item information and the text semantic vector respectively to obtain a first retrieval result and a second retrieval result; and generate context information of the abnormal item information based on the first retrieval result and the second retrieval result.

[0137] In some embodiments, the knowledge base includes document segments of at least one technical document and semantic vectors of corresponding document segments, each technical document includes at least one of a historical inspection report and a product document, and the document segment of each technical document carries dependency information, and the dependency information is used to indicate the hierarchical structure of the corresponding document segment in the corresponding technical document; when the retrieval module 410 is used to search the knowledge base based on the cleaned abnormal item information and the text semantic vector respectively to obtain the first retrieval result and the second retrieval result, it is specifically used for: performing keyword matching in the document segments of each technical document based on the keywords in the cleaned abnormal item information, and taking the matched document segment as the first retrieval result; using the text semantic vector to perform vector retrieval in the semantic vector of the corresponding document segment, and taking the document segment corresponding to the retrieved similar semantic vector as the second retrieval result, and the similarity between the similar semantic vector and the text semantic vector is greater than or equal to the predetermined similarity threshold.

[0138] In some embodiments, when the retrieval module 410 is used to generate context information of abnormal item information based on the first retrieval result and the second retrieval result, it is specifically used to: determine each first weighted similarity based on each first similarity and a preset first weight, and the each first similarity is: the semantic similarity between each document fragment in the first retrieval result and the keyword; determine each second weighted similarity based on each second similarity and a preset second weight, and the each second similarity is: the vector similarity between the semantic vector of each document fragment in the second retrieval result and the text semantic vector; select at least one document fragment with the largest weighted similarity from the first retrieval result and the second retrieval result to generate context information.

[0139] In some embodiments, the summary module 420 is specifically used to merge each document fragment in the context information with the corresponding dependency information to obtain relevant documents of the context information; add the relevant documents of the context information to the abnormal item information to obtain merged abnormal item information; based on the merged abnormal item information, use the large language model to perform abnormal summary of the inspection items to obtain abnormal item summary text.

[0140] In some embodiments, the large language model is a language model that is fine-tuned based on a pre-constructed standard data set, and the standard data set includes: various abnormal scenarios of the detection object and standardized description information of each abnormal scenario; the summary module 420, when used to perform abnormal summary of the inspection items based on the merged abnormal item information using the large language model to obtain the abnormal item summary information, is specifically used to: generate target description information based on the merged abnormal item information using the fine-tuned language model, the target description information is: standardized description information of the abnormal scenario corresponding to the merged abnormal item information; generate abnormal item summary information based on the target description information.

[0141] In some embodiments, the summary module 420, when used to generate abnormal item summary information based on the target description information, is specifically used to: obtain preset prompt words, the prompt words are used to guide the fine-tuned language model to perform cluster analysis of abnormal items of the same type on abnormal scenarios according to the target description information; use the fine-tuned language model, based on the prompt words and the target description information, to perform cluster analysis of abnormal items of the same type on the abnormal scenarios to which the target description information belongs, to obtain standardized description information grouped according to abnormal items of the same type; generate an abnormal item summary text according to the standardized description information grouped according to abnormal items of the same type.

[0142] In some embodiments, the knowledge base includes a problem solution corresponding to standardized description information of at least one abnormal scenario; the device also includes: a solution adding module, which is used to retrieve a problem solution corresponding to the target description information from the knowledge base after updating the inspection report using the abnormal item summary text; and add the retrieved problem solution to the inspection report.

[0143] In some embodiments, the device also includes: a knowledge base construction module, which is used to obtain at least one technical document and problem solution of the detection object before retrieving documents semantically related to the abnormal item information from a predetermined knowledge base based on the abnormal item information in the inspection report, wherein each technical document includes at least one of a historical inspection report and a product document; segmenting each technical document based on the hierarchical structure of each technical document to obtain document segments of each technical document; adding dependency information to the document segment of each technical document, wherein the dependency information is used to indicate the hierarchical structure of the corresponding document segment in the corresponding technical document; vectorizing the document segment with the dependency information added to obtain the semantic vector of the corresponding document segment; generating a knowledge base based on the document segment with the dependency information added, the semantic vector of the corresponding document segment, and the problem solution.

[0144] In some embodiments, the knowledge base construction module is further used to: before adding dependency information to the document slices of each technical document, generate dependency information of the document slices of each technical document according to the hierarchical structure of the document slices of each technical document in the corresponding technical document.

[0145] The functions of each module in each device in the embodiments of the present application can be found in the corresponding description in the above method, and have corresponding beneficial effects, which will not be repeated here.

[0146] Figure 5 The structure diagram of the inspection report generation device provided in the embodiment of the present application is shown. Corresponding to the application scenario and method of the inspection report generation method for cloud network networking provided in the embodiment of the present application, the embodiment of the present application also provides an inspection report generation device for cloud network networking, including: The acquisition module 510 is used to obtain the abnormality detection result of any inspection item obtained by inspecting the cloud network networking product; the information generation module 520 is used to generate abnormal item information according to the abnormality detection result; the retrieval module 530 is used to retrieve documents semantically related to the abnormal item information from a predetermined knowledge base based on the abnormal item information as the context information of the abnormal item information; the summary module 540 uses a predetermined large language model to perform abnormal summary of the inspection items based on the context information and the abnormal item information to obtain the abnormal item summary text; the report generation module 550 is also used to generate an inspection report of the cloud network networking product using the abnormal item summary text.

[0147] In some embodiments, the inspection report includes standardized description information of abnormal scenarios corresponding to the abnormal item information; the device also includes: a solution adding module, which is used to retrieve problem solutions corresponding to the standardized description information from the knowledge base after generating an inspection report for the cloud network networking product using the abnormal item summary text; and add the retrieved problem solutions to the inspection report.

[0148] The functions of each module in each device in the embodiments of the present application can be found in the corresponding description in the above method, and have corresponding beneficial effects, which will not be repeated here.

[0149] Figure 6 FIG. 1 is a block diagram of an electronic device used to implement an embodiment of the present application. Figure 6 As shown, the electronic device includes: a memory 601 and a processor 602, and the memory 601 stores a computer program that can be run on the processor 602. When the processor 602 executes the computer program, the method in the above embodiment is implemented. The number of the memory 601 and the processor 602 can be one or more. In a specific implementation, the electronic device may also include a communication interface 603 for communicating with external devices and performing data exchange transmission.

[0150] In specific implementation, if the memory 601, the processor 602 and the communication interface 603 are implemented independently, the memory 601, the processor 602 and the communication interface 603 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0151] Optionally, in a specific implementation, if the memory 601, the processor 602 and the communication interface 603 are integrated on a chip, the memory 601, the processor 602 and the communication interface 603 can communicate with each other through an internal interface.

[0152] An embodiment of the present application provides a computer-readable storage medium storing a computer program, which implements the method provided in the embodiment of the present application when the program is executed by a processor.

[0153] An embodiment of the present application provides a computer program product, including a computer program, which implements the method provided in the embodiment of the present application when executed by a processor.

[0154] An embodiment of the present application also provides a chip, which includes a processor for calling and executing instructions stored in the memory from the memory, so that a communication device equipped with the chip executes the method provided by the embodiment of the present application.

[0155] An embodiment of the present application also provides a chip, including: an input interface, an output interface, a processor and a memory, wherein the input interface, the output interface, the processor and the memory are connected via an internal connection path, and the processor is used to execute the code in the memory. When the code is executed, the processor is used to execute the method provided in the embodiment of the application.

[0156] It should be understood that the above processor can be a processor (Central Processing Unit, CPU), and can also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application-specific integrated circuits (Application Specific Integrated Circuit, ASIC), field programmable gate arrays (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. It is worth noting that the processor can be a processor that supports the Advanced RISC Machines (ARM) architecture.

[0157] Further, optionally, the above-mentioned memory may include a read-only memory and a random access memory. The memory may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. Among them, the non-volatile memory may include a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may include a random access memory (RAM), which is used as an external cache. By way of exemplary but not limiting description, many forms of RAM are available. For example, static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct memory bus random access memory (DR RAM).

[0158] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium.

[0159] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0160] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0161] Any process or method described in the flow chart or otherwise described herein can be understood as a module, fragment or portion of a code representing one or more executable instructions for implementing the steps of a specific logical function or process. And the scope of the preferred embodiment of the present application includes other implementations, in which the functions may not be performed in the order shown or discussed, including in a substantially simultaneous manner or in a reverse order according to the functions involved.

[0162] The logic and / or steps described in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by an instruction execution system, apparatus or device (such as a computer-based system, a system including a processor or other system that can fetch instructions from an instruction execution system, apparatus or device and execute instructions), or used in combination with these instruction execution systems, apparatuses or devices.

[0163] It should be understood that the various parts of the present application can be implemented with hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented with software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the above embodiment method can be completed by instructing the relevant hardware through a program, which can be stored in a computer-readable storage medium, and when the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0164] In addition, each functional unit in each embodiment of the present application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into one module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of a software functional module. If the above-mentioned integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk or an optical disk, etc.

[0165] The above is only an exemplary embodiment of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various changes or substitutions within the technical scope recorded in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for generating an inspection report in a cloud network networking scenario, characterized in that: The method comprises: Based on the acquired abnormal item information, documents semantically related to the abnormal item information are retrieved from a predetermined knowledge base as context information of the abnormal item information; the abnormal item information is used to indicate an abnormal detection result of any inspection item of a detection object, and the detection object is one or more products used for cloud network networking; Based on the context information and the abnormal item information, using a predetermined large language model to perform abnormal summary on the inspection item to obtain an abnormal item summary text; An inspection report is generated using the abnormal item summary text.

2. The method according to claim 1, characterized in that The retrieving documents semantically related to the abnormal item information from a predetermined knowledge base as context information of the abnormal item information includes: Performing data cleaning on the abnormal item information to obtain cleaned abnormal item information; Vectorizing the cleaned abnormal item information to obtain a text semantic vector; Based on the cleaned abnormal item information and the text semantic vector, the knowledge base is searched respectively to obtain a first search result and a second search result; Context information of the abnormal item information is generated according to the first search result and the second search result.

3. The method according to claim 2, characterized in that The knowledge base includes at least one document segment of a technical document and a semantic vector of the corresponding document segment, each technical document includes at least one of a historical inspection report and a product document, and each document segment of the technical document carries dependency information, and the dependency information is used to indicate the hierarchical structure of the corresponding document segment in the corresponding technical document; The searching the knowledge base based on the cleaned abnormal item information and the text semantic vector to obtain a first search result and a second search result includes: Perform keyword matching in the document segments of each technical document based on the keywords in the cleaned abnormal item information, and use the matched document segments as the first search result; The text semantic vector is used to perform vector retrieval in the semantic vector of the corresponding document segment, and the document segment corresponding to the retrieved similar semantic vector is used as the second retrieval result, and the similarity between the similar semantic vector and the text semantic vector is greater than or equal to a predetermined similarity threshold.

4. The method according to claim 3, characterized in that The generating, according to the first search result and the second search result, context information of the abnormal item information includes: Determine each first weighted similarity based on each first similarity and a preset first weight, wherein each first similarity is: a semantic similarity between each document fragment in the first search result and the keyword; Based on each second similarity and a preset second weight, determining each second weighted similarity, wherein each second similarity is: a vector similarity between a semantic vector of each document segment in the second search result and the text semantic vector; At least one document fragment with the greatest weighted similarity is selected from the first retrieval result and the second retrieval result to generate the context information.

5. The method according to claim 3, characterized in that: The abnormal summary of the inspection items is performed based on the context information and the abnormal item information by using a predetermined large language model to obtain an abnormal item summary text, including: Merging each document fragment in the context information with the corresponding dependency information to obtain a document related to the context information; Adding the relevant documents of the context information to the abnormal item information to obtain the merged abnormal item information; Based on the merged abnormal item information, the inspection items are abnormally summarized using the large language model to obtain an abnormal item summary text.

6. The method according to claim 5, characterized in that The large language model is a language model fine-tuned based on a pre-built standard data set, wherein the standard data set includes: each abnormal scene of the detection object and standardized description information of each abnormal scene; Based on the merged abnormal item information, using the large language model to perform abnormal summary on the inspection items to obtain abnormal item summary information includes: Generate target description information based on the merged abnormal item information using the fine-tuned language model, wherein the target description information is: standardized description information of the abnormal scenario corresponding to the merged abnormal item information; The abnormal item summary information is generated according to the target description information.

7. The method according to claim 6, characterized in that The generating the abnormal item summary information according to the target description information includes: Obtaining a preset prompt word, wherein the prompt word is used to guide the fine-tuned language model to perform cluster analysis of abnormal items of the same type on the abnormal scene according to the target description information; Using the fine-tuned language model, based on the prompt word and the target description information, cluster analysis of abnormal items of the same type is performed on the abnormal scene to which the target description information belongs, so as to obtain standardized description information grouped according to abnormal items of the same type; The abnormal item summary text is generated according to the standardized description information grouped according to the abnormal items of the same type.

8. The method according to claim 7, characterized in that The knowledge base includes a problem solution corresponding to standardized description information of at least one abnormal scenario; after the inspection report is updated using the abnormal item summary text, it also includes: Retrieving a problem solution corresponding to the target description information from the knowledge base; The retrieved problem solution is added to the inspection report.

9. The method according to claim 3, characterized in that: Before retrieving documents semantically related to the abnormal item information from a predetermined knowledge base based on the abnormal item information in the inspection report, the method further includes: Obtain at least one technical document and a problem solution for the inspection object, each technical document including at least one of a historical inspection report and a product document; Slicing each technical document based on its hierarchical structure to obtain document slices of each technical document; Adding the dependency information to the document segments of each technical document, and performing vectorization processing on the document segments to which the dependency information is added, to obtain semantic vectors of the corresponding document segments; The knowledge base is generated according to the document segments to which the dependency information is added, the semantic vectors corresponding to the document segments and the problem solution.

10. The method according to claim 9, characterized in that Before adding the dependency information to the document segments of each technical document, the method further includes: Dependency information of the document segments of each technical document is generated according to the hierarchical structure of the document segments of each technical document in the corresponding technical document.

11. A method for generating an inspection report for a cloud network, characterized in that: The method comprises: Obtain the abnormal detection result of any inspection item from the inspection results of cloud network networking products; generating abnormal item information according to the abnormality detection result; Based on the abnormal item information, retrieve documents semantically related to the abnormal item information from a predetermined knowledge base as context information of the abnormal item information; Based on the context information and the abnormal item information, using a predetermined large language model to perform abnormal summary on the inspection item to obtain an abnormal item summary text; The abnormal item summary text is used to generate an inspection report of the cloud network networking product.

12. The method according to claim 11, characterized in that The inspection report includes standardized description information of the abnormal scenario corresponding to the abnormal item information; after the inspection report of the cloud network networking product is generated using the abnormal item summary text, it also includes: Retrieving a problem solution corresponding to the standardized description information from the knowledge base; The retrieved problem solution is added to the inspection report.

13. An electronic device comprising a memory, a processor and a computer program stored in the memory, wherein the processor implements the method of any one of claims 1 to 10 or any one of claims 11-12 when executing the computer program.

14. A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the method according to any one of claims 1 to 10 or any one of claims 11-12 is implemented.

15. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 10 or any one of claims 11-12.

Images