Work order clustering and theme extraction method, system, equipment and medium

Abstract

The invention provides a work order clustering and theme extraction method, system and device and a medium, and belongs to the technical field of natural language processing and data mining. The method comprises the steps of obtaining original work order data, performing data cleaning, extracting work order abstracts from the original work order data by using a text abstract generation technology according to preset abstract constraints, and generating an abstract set; inputting the work order abstract into a language model, generating a semantic vector of a preset dimension, constructing a vector index through an FAISS library, and converting the abstract set into a set of numerical vectors; clustering the numerical vectors by using an improved K-means algorithm, and outputting K groups of work order digests with balanced quantity; for each group of work order abstracts, screening representative abstracts based on text similarity; extracting keywords from the representative abstracts by adopting a keyword extraction technology, and generating a keyword list of each group of work order abstracts; and for the keyword list of each group of work order abstracts, generating a theme tag by combining keywords.

Description

Technical Field

[0001] This invention belongs to the field of natural language processing and data mining technology, and more specifically relates to a work order clustering and topic extraction method, system, device and medium. Background Technology

[0002] With the continuous advancement of enterprise informatization, work order systems, such as IT service management systems and customer complaint systems, are playing an increasingly important role in daily enterprise operations, resulting in an explosive growth in the amount of data generated. Taking a large enterprise as an example, its work order system receives up to a million new work orders daily. These work orders contain various fields such as request details, submission time, and priority, containing a wealth of business information. However, efficiently processing such massive amounts of data has become a pressing problem for enterprises. Currently, traditional work order processing methods mainly rely on manual annotation or simple keyword matching. While manual annotation can ensure classification accuracy to a certain extent, it is extremely costly and inefficient when dealing with millions of work orders, making it difficult to meet the real-time processing needs of enterprises. Simple keyword matching methods can achieve preliminary classification, but they suffer from serious semantic understanding deficiencies, failing to capture text semantic similarity and often leading to the misclassification of similar work orders, affecting the targetedness and effectiveness of subsequent work order processing. Furthermore, both of these traditional methods have extremely high requirements for memory and computing resources when processing large-scale data, making it difficult to complete the text processing task of millions of work orders in a single-machine environment. Data scale has become a key factor restricting their application. In recent years, deep learning and clustering algorithms have made significant progress in text processing, bringing new ideas to work order processing. However, when these technologies are directly applied to large-scale work order data, they still face many challenges. In the text vectorization stage, traditional methods such as TF-IDF and Word2Vec are difficult to effectively process the semantic information of long work order texts and cannot accurately extract the core semantics of the text. In the clustering stage, high-dimensional vector clustering algorithms suffer from high computational complexity and large memory consumption, making them difficult to adapt to the real-time clustering requirements of large-scale data. In terms of topic extraction, existing technologies often rely on a lot of manual intervention, with low automation, making it difficult to quickly and comprehensively cover all work order categories and meet the actual needs of enterprises for efficient work order processing. In summary, existing work order processing technologies suffer from shortcomings such as insufficient data processing capabilities, semantic understanding biases, and low efficiency in topic extraction when dealing with large-scale work order data. There is an urgent need for a lightweight and automated work order clustering and topic extraction method to achieve efficient processing of large-scale work order data. Summary of the Invention

[0003] To address the above problems, the present invention aims to provide a method, system, device, and medium for work order clustering and topic extraction. Through intelligent adaptive summary extraction algorithms, lightweight semantic vector processing, dynamic balanced clustering, multi-dimensional keyword extraction, and structured topic generation, it achieves efficient processing and accurate semantic extraction of work order data, significantly improving the intelligence level of work order classification management and business analysis efficiency.

[0004] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, embodiments of this application provide a method for work order clustering and topic extraction, including: After obtaining the original work order data and cleaning it, the TextRank algorithm or BART model is used to extract the work order summary from the original work order data according to the preset summary constraints, and a summary set is generated. Read the summary set, input the work order summary into a lightweight pre-trained language model, generate semantic vectors of a preset dimension, and build a vector index through the FAISS library to convert the summary set into a set of numerical vectors; Read the set of numerical vectors, apply the improved K-means algorithm to cluster the numerical vectors, and output K groups of work order summaries with balanced number of orders; For each set of work order summaries, representative summaries are selected based on text similarity. Keywords are extracted from representative summaries using the TF-IDF algorithm or the BERTopic model to generate a keyword list for each work order summary. For each set of work order summaries, a keyword list is generated by merging the keywords to create topic tags.

[0005] In an optional implementation, after acquiring the original work order data and performing data cleaning, the textRank algorithm or BART model is used to extract work order summaries from the original work order data according to preset summary constraints, and a summary set is generated, including: Obtain the original work order data, encode the original work order data into UTF-8, remove special symbols, redundant spaces, newlines and meaningless characters, and delete meaningless words based on the general stop word list and business-defined stop words. Check whether there are labeled data in the processed original work order data; If so, then based on the preset summary text length, a supervised BART pre-trained summarization model is used to extract work order summaries from the original work order data, and a summary set is generated; if not, based on the preset summary text length, an unsupervised TextRank algorithm is used to extract work order summaries from the original work order data, and a summary set is generated.

[0006] In an optional implementation, the step of reading the summary set involves inputting the work order summary into a lightweight pre-trained language model to generate semantic vectors of a preset dimension, and constructing a vector index using the FAISS library to convert the summary set into a set of numerical vectors, including: Read the summary set, input the work order summary into the Sentence–BERT model or MiniLM model, and generate a 768-dimensional semantic vector; For a 768-dimensional semantic vector, the 32-bit floating-point number is compressed into an 8-bit integer; The FAISS library is used to construct a vector index, and the semantic vectors are normalized using the L2 norm to generate a set of numerical vectors.

[0007] In an optional implementation, the set of read numerical vectors is clustered using an improved K-means algorithm to output K groups of work order summaries with balanced numbers, including: The K-means++ algorithm is used to select the initial cluster centers; During the iteration phase, the numerical vectors are processed in batches based on the set of numerical vectors. 100,000 numerical vectors are loaded each time. The distance between the numerical vector and the cluster center is calculated using cosine similarity as the distance metric. The number of clusters K is dynamically adjusted according to the elbow rule or silhouette coefficient. Finally, K groups of work order summaries are output.

[0008] In an optional implementation, the step of screening representative summaries based on text similarity for each group of work order summaries includes: For each group of work order summaries, calculate the cosine similarity between the work order summaries within the group and the text cluster centers, and retain the N work order summaries with the highest similarity as representative summaries.

[0009] In an optional implementation, the step of extracting keywords from representative summaries using the TF-IDF algorithm or the BERTopic model to generate a keyword list for each work order summary includes: When using the TF-IDF algorithm, the TF-IDF algorithm is used to statistically analyze the term frequency and inverse document frequency in the representative summary, and keywords are screened out according to the preset frequency threshold to generate a keyword list. When using the BERTopic model, a representative summary is input into the BERTopic model to generate a summary semantic cluster; the core words within the summary semantic cluster are extracted as keywords to generate a keyword list.

[0010] In an optional implementation, the keyword list for each set of work order summaries is used to generate topic tags by merging keywords, including: For each work order summary, extract the keywords from the keyword list, identify keywords of the same type, and merge them according to preset rules. Input the processed keywords into a predefined template to generate structured tags; Identify semantically repetitive words in structured tags, merge them, delete meaningless tags from structured tags, and generate topic tags.

[0011] Secondly, embodiments of this application also provide a work order clustering and topic extraction system, including: The summary extraction module is used to acquire raw work order data, perform data cleaning, and extract work order summaries from the raw work order data using the TextRank algorithm or BART model according to preset summary constraints, and generate a summary set. The text vectorization module is used to read the summary set, input the work order summary into a lightweight pre-trained language model, generate semantic vectors of a preset dimension, and build a vector index through the FAISS library to convert the summary set into a set of numerical vectors. The summary clustering module is used to read a set of numerical vectors, apply an improved K-means algorithm to cluster the numerical vectors, and output K groups of work order summaries with balanced numbers. The summary filtering module is used to filter out representative summaries for each group of work order summaries based on text similarity; The keyword extraction module is used to extract keywords from representative summaries using the TF-IDF algorithm or the BERTopic model, and generate a keyword list for each work order summary. The topic tag generation module is used to generate topic tags by merging keywords from the keyword list for each work order summary.

[0012] Thirdly, embodiments of this application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the work order clustering and topic extraction method described in any of the above descriptions.

[0013] Fourthly, embodiments of this application also provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the work order clustering and topic extraction steps as described in any of the above.

[0014] As can be seen from the above technical solutions, the present invention has the following advantages: The work order clustering and topic extraction method provided in this application removes redundant information through data cleaning during the data processing stage. It flexibly selects either the BART model or the TextRank algorithm to extract work order summaries based on whether the data is labeled or not, ensuring the accuracy and adaptability of the summary extraction. Lightweight pre-trained language models such as Sentence-BERT and MiniLM are used to generate semantic vectors and compress them. A vector index is constructed using the FAISS library, reducing computational costs while improving vector processing efficiency. An improved K-means algorithm is employed, utilizing K-means++ to select initial cluster centers, process numerical vectors in batches, and dynamically adjust the number of clusters K to achieve balanced clustering of work order summaries and improve clustering performance. Representative summaries are screened based on text similarity, and keywords are extracted using the TF-IDF algorithm or the BERTopic model, enabling accurate extraction of key information from different perspectives. By merging keywords to generate topic tags, and after structured processing and deduplication, concise, accurate, and meaningful topic tags are generated, providing strong support for work order classification management and business analysis, and improving the intelligence level and efficiency of work order processing.

[0015] This application removes redundant information through data cleaning and dynamically selects either the BART model (supervised) or the TextRank algorithm (unsupervised) to extract work order summaries based on the data labeling status, ensuring the accuracy and scenario adaptability of the summary extraction and laying a high-quality data foundation for subsequent processing.

[0016] This application uses lightweight models such as Sentence-BERT / MiniLM to generate semantic vectors, and combines 32-bit floating-point compression technology with FAISS library vector indexing to reduce computational resource consumption while improving vector processing efficiency, thus achieving efficient semantic vectorization conversion.

[0017] This application optimizes the initial cluster center selection based on K-means++. By processing large-scale data in batches (100,000 records each time) and dynamically adjusting the number of clusters K in combination with the elbow rule / profile coefficient, it achieves balanced clustering of work order summaries, significantly improving the stability and rationality of clustering results.

[0018] This application selects representative abstracts through text similarity and flexibly uses the TF-IDF algorithm (statistical word frequency feature) or BERTopic model (semantic cluster analysis) to extract keywords, accurately capturing core information from two dimensions: statistical features and semantic depth, covering different business analysis needs.

[0019] This application merges keyword lists of the same type, performs template structuring and semantic deduplication to generate concise and accurate topic tags, providing a standardized semantic tagging system for work order classification management and business trend analysis, effectively improving the intelligence level and work efficiency of work order processing. Attached Figure Description

[0020] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating the work order clustering and topic extraction method provided in this application.

[0022] Figure 2 This is a flowchart illustrating the process of grouping work order summaries using a lightweight clustering algorithm, as provided in this application.

[0023] Figure 3 This is a schematic diagram of the work order clustering and topic extraction system provided in this application.

[0024] Figure 4 A schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0025] The various embodiments of this disclosure will be described more fully in the detailed steps of the work order clustering and topic extraction method described below. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0026] In the following, the terms “comprising” or “may include”, which may be used in various embodiments of this disclosure, indicate the presence of the disclosed functions, operations, or elements, and do not limit the addition of one or more functions, operations, or elements. Furthermore, as used in various embodiments of this disclosure, the terms “comprising,” “having,” and their cognates are intended only to indicate a particular feature, number, step, operation, element, component, or combination of the foregoing, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing, or the possibility of adding one or more combinations of the foregoing.

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Please refer to Figure 1 The following is a method flowchart of a work order clustering and theme extraction method in a specific embodiment. The method includes: S1: Obtain the original work order data. After data cleaning, use the TextRank algorithm or the BART model according to the preset summary constraint to extract the work order summary from the original work order data, and generate a summary set.

[0029] In the specific implementation, first, obtain the original work order data, encode the text of the original work order data as UTF-8, and remove garbled characters and invisible characters; remove special symbols, redundant spaces, line breaks and meaningless characters (such as pure numbers, repeated punctuation) therein, and delete meaningless words based on the general stop word list (such as Chinese words like "de", "le", "he") and business custom stop words.

[0030] Then, identify whether there is labeled data in the processed original work order data; If so, according to the preset summary text length, use the supervised BART pre-trained summary model to extract the work order summary from the original work order data, and generate a summary set. Among them, the supervised BART pre-trained summary model can be fine-tuned to adapt to the work order domain text and has the advantage of high summary quality.

[0031] If not, according to the preset summary text length, use the unsupervised TextRank algorithm to extract the work order summary from the original work order data, and generate a summary set. Among them, the unsupervised TextRank algorithm is suitable for unlabeled data and has the advantage of high computational efficiency.

[0032] In this step, the preset summary text length is set to be less than or equal to 50 characters, which can ensure semantic retention. In addition, manual sampling verification can be performed (such as extracting 1000 for comparison with the original text), and it is required that the summary coverage rate ≥ 85%.

[0033] Through this step, the memory occupancy of the work order data set after cleaning and summarization is reduced by about 90% compared with the original data (for example, from 1GB compressed to 100MB), significantly reducing the resource requirements for subsequent processing.

[0034] It can be seen that through the above steps, the original work order data is converted into a structured and low-redundancy summary set, laying a foundation for efficient text vectorization and clustering.

[0035] S2: Read the summary set, input the work order summary into the lightweight pre-trained language model to generate semantic vectors of a preset dimension, and construct a vector index through the FAISS library to convert the summary set into a set of numerical vectors.

[0036] In specific implementations, a lightweight pre-trained language model is first selected, such as Sentence-BERT or MiniLM. These models are based on the Transformer architecture and learn general semantic representations through pre-training on massive amounts of text, enabling them to capture deep semantic information of the text. When selecting a model, a balance must be struck between accuracy and computational efficiency: for example, the MiniLM model has a smaller parameter scale (e.g., a 6-layer Transformer), but its semantic expressive power is close to that of BERT-base, making it suitable for resource-constrained scenarios.

[0037] After selecting a model, the summary set is read, and the work order summary is input into the Sentence-BERT model or the MiniLM model to generate a 768-dimensional semantic vector. Here, 768 dimensions is the common output dimension of BERT-like models, achieving a good balance between semantic expressiveness and computational complexity.

[0038] Then, to further reduce memory usage, the generated floating-point vectors of the 768-dimensional semantic vectors are quantized, that is, the 32-bit floating-point numbers are compressed into 8-bit integers, reducing the vector storage space by 75%, while maintaining semantic accuracy through model fine-tuning.

[0039] Finally, the FAISS library is used to construct a vector index, and the semantic vectors are normalized using the L2 norm to generate a set of numerical vectors.

[0040] The FAISS (FacebookAISimilaritySearch) library is used to build a vector index, supporting efficient similarity search. Vector normalization (L2 norm normalization) ensures the stability of cosine similarity calculation.

[0041] This step efficiently converts work order summaries into low-dimensional semantic vectors, providing high-quality input for subsequent clustering while keeping memory usage within a reasonable range.

[0042] S3: Read the set of numerical vectors, apply the improved K-means algorithm to cluster the numerical vectors, and output K groups of work order summaries with balanced numbers.

[0043] In a specific implementation, the K-means++ algorithm is used to select the initial cluster centers. During the iteration phase, the numerical vectors are processed in batches based on the set of numerical vectors, with 100,000 numerical vectors loaded each time. The distance between the numerical vectors and the cluster centers is calculated using cosine similarity as the distance metric. The number of clusters K is dynamically adjusted according to the elbow rule or silhouette coefficient, and finally, K groups of work order summaries are output.

[0044] For example, this step uses a lightweight clustering algorithm to group work order summaries, such as... Figure 2 As shown, the specific process includes: a) Initialization phase: The K-means++ algorithm is used to select initial cluster centers to avoid local optima; b) Iteration phase: The vectors are processed in batches, loading 100,000 vectors into memory at a time. Cosine similarity is used as a distance metric to calculate the distance between a vector and the cluster center. The number of clusters K is dynamically adjusted (optimized using the elbow rule or silhouette coefficient).

[0045] c) Output: K groups of similar work orders, with a balanced number of work orders in each group (error ≤ 5%).

[0046] S4: For each work order summary, select representative summaries based on text similarity.

[0047] In a specific implementation, for each group of work order summaries, the cosine similarity between the work order summaries within the group and the text cluster center is calculated, and the N work order summaries with the highest similarity are retained as representative summaries.

[0048] For example, for each clustered work order summary, the top N (N=5~10) representative summaries can be filtered by text similarity or frequency of occurrence.

[0049] When using text similarity, the cosine similarity between the summary within a group and the cluster center is calculated, and the N most similar items are retained. When using frequency filtering, high-frequency words or phrases in the statistical abstracts are counted, and the N abstracts that cover the most core semantics are selected as representative abstracts.

[0050] S5: Use the TF-IDF algorithm or BERTopic model to extract keywords from representative summaries and generate a keyword list for each work order summary.

[0051] In specific implementations, two methods can be used for keyword extraction: When using the TF-IDF algorithm, the TF-IDF algorithm is used to count the term frequency and inverse document frequency in representative summaries. Based on the preset frequency threshold, significant words in the summaries within the group are selected as keywords to generate a keyword list.

[0052] When using the BERTopic model, a representative summary is input into the BERTopic model to generate a summary semantic cluster; the core words within the summary semantic cluster are extracted as keywords to generate a keyword list.

[0053] S6: For each set of work order summaries, generate topic tags by merging the keyword list.

[0054] In a specific implementation, the keywords are first extracted from the keyword list of each work order summary, and similar keywords are identified and merged according to preset rules. For example, similar keywords are combined, such as "network failure" + "network outage" combined into "network failure - network outage".

[0055] Then, input the processed keywords into a predefined template (such as "problem type - specific scenario") to generate structured tags. At this point, low-confidence tags can be manually verified to ensure that the topic conforms to the business logic.

[0056] Finally, semantically repetitive words in the structured tags (such as "network interruption" and "network disconnection") are identified, merged, and meaningless tags (such as "other" and "unknown") are deleted to ensure topic coverage ≥90% and generate topic tags.

[0057] Through the above steps, each work order is refined into concise and semantically clear topic tags, significantly reducing the cost of manual review and improving the automation level of work order classification. In this embodiment, the efficiency and quality of work order topic analysis are significantly improved through an end-to-end automated process: First, in the data preprocessing stage, a hybrid summary extraction strategy (unsupervised TextRank and supervised BART model) is adopted to flexibly adapt to different data scenarios, and the cleaning process effectively eliminates noise interference, laying a high-quality input foundation for subsequent processing; then, a lightweight pre-trained model (such as Sentence-BERT / MiniLM) is used to generate semantic vectors, and the FAISS library is used to achieve vector compression and fast retrieval, which greatly reduces the consumption of computing resources while ensuring the accuracy of semantic representation; In the clustering stage, the improved K-means algorithm introduces a dynamic clustering quantity adjustment mechanism and a batch processing strategy, balancing large-scale data processing capabilities with clustering effect optimization. Furthermore, representative summaries within each group are selected using cosine similarity, and core keywords are extracted using TF-IDF or BERTopic models, forming a multi-level semantic filtering mechanism. Finally, based on keyword merging and templated tag generation technology, semantic redundancy is automatically eliminated and structured topic tags are output. The entire process achieves efficient transformation from raw work orders to interpretable topics, demonstrating comprehensive advantages in processing speed, resource utilization, and topic representation accuracy.

[0058] like Figure 3 As shown below, the following are embodiments of the work order clustering and topic extraction system provided in this disclosure. This system and the work order clustering and topic extraction methods in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the work order clustering and topic extraction system, please refer to the embodiments of the above work order clustering and topic extraction methods.

[0059] A work order clustering and topic extraction system includes: The summary extraction module is used to acquire raw work order data, perform data cleaning, and extract work order summaries from the raw work order data using the TextRank algorithm or BART model according to preset summary constraints, and generate a summary set.

[0060] The text vectorization module is used to read the summary set, input the work order summary into a lightweight pre-trained language model, generate semantic vectors of preset dimensions, and build a vector index through the FAISS library to convert the summary set into a set of numerical vectors.

[0061] The summary clustering module is used to read a set of numerical vectors, apply an improved K-means algorithm to cluster the numerical vectors, and output K groups of work order summaries with balanced numbers.

[0062] The summary filtering module is used to filter out representative summaries for each group of work order summaries based on text similarity.

[0063] The keyword extraction module is used to extract keywords from representative summaries using the TF-IDF algorithm or the BERTopic model, generating a keyword list for each work order summary.

[0064] The topic tag generation module is used to generate topic tags by merging keywords from the keyword list for each work order summary.

[0065] The work order clustering and topic extraction system provided in this embodiment achieves efficient processing and accurate topic extraction of work order data by constructing an end-to-end automated analysis process. First, a hybrid summary extraction strategy flexibly adapts to different data scenarios, and the cleaning process ensures input quality. Subsequently, a lightweight semantic model and FAISS vector retrieval technology significantly reduce computational resource consumption while maintaining high-dimensional semantic representation capabilities. The improved clustering algorithm balances the efficiency of large-scale data processing and clustering accuracy through dynamic adjustment of the number of clusters and batch processing mechanisms. A multi-level semantic filtering mechanism (similarity screening + keyword extraction) further enhances the representativeness of topics. Finally, structured topic tags are generated through intelligent merging and template technology. The overall solution achieves comprehensive optimization in processing speed, resource utilization, and topic interpretability, effectively supporting the needs of work order classification and business insight.

[0066] Figure 4 A schematic diagram of the hardware structure of an electronic device for implementing various embodiments of the present invention.

[0067] The work order clustering and topic extraction method provided in this application can be applied to electronic devices. Those skilled in the art will understand that the electronic device structure involved in the embodiments of this invention does not constitute a limitation on the electronic device. An electronic device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, electronic devices include, but are not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0068] Electronic devices may include processors, external memory interfaces, internal memory, universal serial bus (USB) interfaces, charging management modules, power management modules, batteries, wireless communication modules, audio modules, speakers, microphones, sensor modules, buttons, cameras, displays, and SIM card interfaces, etc.

[0069] A processor may include one or more processing units, such as: a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0070] The processor can serve as the nerve center and command center of an electronic device. The controller can generate operation control signals based on the instruction opcode and timing signals to control the fetching and execution of instructions.

[0071] The processor may also include memory for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can store instructions or data that the processor has just used or that are used repeatedly. If the processor needs to use the instruction or data again, it can retrieve it directly from this memory. This avoids repeated accesses, reduces processor latency, and thus improves system efficiency.

[0072] An external storage interface (ESI) can be used to connect external memory cards, such as microSD cards, to expand the storage capacity of electronic devices. The external memory card communicates with the processor through the ESI to perform data storage functions, such as saving music and video files on the external memory card.

[0073] Internal memory can be used to store computer executable program code, which includes instructions. The processor executes various functional applications and data processing of electronic devices by running the instructions stored in internal memory. Internal memory can include a program storage area and a data storage area. Internal memory can include high-speed random access memory, and can also include non-volatile memory, such as at least one disk storage device, flash memory device, universal flash storage (UFS), etc.

[0074] Wireless communication functionality in electronic devices can be achieved through antennas, wireless communication modules, modem processors, and baseband processors.

[0075] Wireless communication modules can provide solutions for wireless communication applications in electronic devices, including wireless local area networks (WLANs) (such as wireless fidelity (Wi-Fi) networks), Bluetooth (BT), global navigation satellite system (GNSS), frequency modulation (FM), near field communication (NFC), and infrared (IR) technologies.

[0076] Electronic devices can implement audio functions through audio modules, speakers, receivers, microphones, headphone jacks, and application processors.

[0077] Electronic devices can achieve shooting functions through ISPs, cameras, video codecs, GPUs, displays, and application processors.

[0078] Electronic devices can achieve display functions through GPUs, displays, and application processors.

[0079] A GPU is a microprocessor for image processing, connected to the display screen and application processor. GPUs are used to perform mathematical and geometric calculations for graphics rendering. A processor may include one or more GPUs, which execute program instructions to generate or modify display information.

[0080] A display screen is used to display images, videos, etc. A display screen includes a display panel.

[0081] The aforementioned electronic device realizes the entire process of the work order clustering and topic extraction method of this application, which involves cleaning the original work order data, intelligently selecting the TextRank or BART model to extract summaries based on the annotation status, generating and compressing semantic vectors using lightweight models such as Sentence-BERT / MiniLM, achieving balanced clustering with the help of the improved K-means algorithm, selecting representative summaries based on text similarity and extracting keywords by combining TF-IDF / BERTopic, and finally merging keywords to generate structured topic tags. This achieves the beneficial effect of improving the accuracy of work order summary extraction, cluster balance, and topic tag generation accuracy while reducing computational costs, providing efficient semantic support for work order classification management and business analysis, and significantly improving the level of intelligent processing.

[0082] The storage medium provided in this application stores a program product capable of implementing work order clustering and topic extraction methods.

[0083] Work order clustering and topic extraction methods include: After obtaining the original work order data and cleaning it, the TextRank algorithm or BART model is used to extract the work order summary from the original work order data according to the preset summary constraints, and a summary set is generated. Read the summary set, input the work order summary into a lightweight pre-trained language model, generate semantic vectors of a preset dimension, and build a vector index through the FAISS library to convert the summary set into a set of numerical vectors; Read the set of numerical vectors, apply the improved K-means algorithm to cluster the numerical vectors, and output K groups of work order summaries with balanced number of orders; For each set of work order summaries, representative summaries are selected based on text similarity. Keywords are extracted from representative summaries using the TF-IDF algorithm or the BERTopic model to generate a keyword list for each work order summary. For each set of work order summaries, a keyword list is generated by merging the keywords to create topic tags.

[0084] In some possible implementations, the work order clustering and topic extraction method of this disclosure can be implemented as a program product, which includes program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure.

[0085] The storage medium disclosed herein may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0086] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for work order clustering and topic extraction, characterized in that, include: After obtaining the original work order data and cleaning it, the TextRank algorithm or BART model is used to extract the work order summary from the original work order data according to the preset summary constraints, and a summary set is generated. Read the summary set, input the work order summary into a lightweight pre-trained language model, generate semantic vectors of a preset dimension, and build a vector index through the FAISS library to convert the summary set into a set of numerical vectors; Read the set of numerical vectors, apply the improved K-means algorithm to cluster the numerical vectors, and output K groups of work order summaries with balanced numbers; For each set of work order summaries, representative summaries are selected based on text similarity. Keywords are extracted from representative summaries using the TF-IDF algorithm or the BERTopic model to generate a keyword list for each work order summary. For each set of work order summaries, a keyword list is generated by merging the keywords to create topic tags.

2. The work order clustering and topic extraction method according to claim 1, characterized in that, After obtaining and cleaning the original work order data, the TextRank algorithm or BART model is used to extract work order summaries from the original work order data according to preset summary constraints, and a summary set is generated, including: Obtain the original work order data, encode the original work order data into UTF-8, remove special symbols, redundant spaces, newlines and meaningless characters, and delete meaningless words based on the general stop word list and business-defined stop words. Check whether there is labeled data in the processed original work order data; If so, then based on the preset summary text length, a supervised BART pre-trained summarization model is used to extract work order summaries from the original work order data, and a summary set is generated; if not, based on the preset summary text length, an unsupervised TextRank algorithm is used to extract work order summaries from the original work order data, and a summary set is generated.

3. The work order clustering and topic extraction method according to claim 2, characterized in that, The process of reading the summary set involves inputting the work order summary into a lightweight pre-trained language model to generate semantic vectors of a preset dimension. A vector index is then constructed using the FAISS library to convert the summary set into a set of numerical vectors, including: Read the summary set, input the work order summary into the Sentence–BERT model or MiniLM model, and generate a 768-dimensional semantic vector; For a 768-dimensional semantic vector, the 32-bit floating-point number is compressed into an 8-bit integer; The FAISS library is used to construct a vector index, and the semantic vectors are normalized using the L2 norm to generate a set of numerical vectors.

4. The work order clustering and topic extraction method according to claim 3, characterized in that, The set of read numerical vectors is clustered using an improved K-means algorithm, outputting K groups of work order summaries with balanced numbers, including: The K-means++ algorithm is used to select the initial cluster centers; During the iteration phase, the numerical vectors are processed in batches based on the set of numerical vectors. 100,000 numerical vectors are loaded each time. The distance between the numerical vector and the cluster center is calculated using cosine similarity as the distance metric. The number of clusters K is dynamically adjusted according to the elbow rule or silhouette coefficient. Finally, K groups of work order summaries are output.

5. The work order clustering and topic extraction method according to claim 4, characterized in that, For each work order summary, representative summaries are selected based on text similarity, including: For each group of work order summaries, calculate the cosine similarity between the work order summaries within the group and the text cluster centers, and retain the N work order summaries with the highest similarity as representative summaries.

6. The work order clustering and topic extraction method according to claim 5, characterized in that, The process involves extracting keywords from representative summaries using the TF-IDF algorithm or the BERTopic model to generate a keyword list for each work order summary, including: When using the TF-IDF algorithm, the TF-IDF algorithm is used to statistically analyze the term frequency and inverse document frequency in the representative summary, and keywords are screened out according to the preset frequency threshold to generate a keyword list. When using the BERTopic model, a representative summary is input into the BERTopic model to generate a summary semantic cluster; the core words within the summary semantic cluster are extracted as keywords to generate a keyword list.

7. The work order clustering and topic extraction method according to claim 6, characterized in that, The keyword list for each work order summary is generated by merging keywords to create topic tags, including: For each work order summary, extract the keywords from the keyword list, identify keywords of the same type, and merge them according to preset rules. Input the processed keywords into a predefined template to generate structured tags; Identify semantically repetitive words in structured tags, merge them, delete meaningless tags from structured tags, and generate topic tags.

8. A work order clustering and topic extraction system, characterized in that, The system employs the work order clustering and topic extraction method as described in any one of claims 1 to 7; The system includes: The abstract extraction module is used to acquire raw work order data, perform data cleaning, and extract work order abstracts from the raw work order data using the TextRank algorithm or BART model according to preset abstract constraints, and generate an abstract set. The text vectorization module is used to read the summary set, input the work order summary into a lightweight pre-trained language model, generate semantic vectors of a preset dimension, and build a vector index through the FAISS library to convert the summary set into a set of numerical vectors. The summary clustering module is used to read a set of numerical vectors, apply an improved K-means algorithm to cluster the numerical vectors, and output K groups of work order summaries with balanced numbers. The summary filtering module is used to filter out representative summaries for each group of work order summaries based on text similarity; The keyword extraction module is used to extract keywords from representative summaries using the TF-IDF algorithm or the BERTopic model, and generate a keyword list for each work order summary. The topic tag generation module is used to generate topic tags by merging keywords from the keyword list for each work order summary.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the work order clustering and topic extraction method as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the work order clustering and topic extraction method as described in any one of claims 1 to 7.

Images