Regulation and retrieval system and method based on word family recommendation

Abstract

The invention discloses a regulation and control retrieval system and method based on word family recommendation, and the system comprises an obtaining module which is used for obtaining preprocessed text data in the power field; the construction module is used for classifying the power field lexicon based on the word frequency to obtain a power field professional lexicon; the analysis module is used for constructing an electric power field word family according to the theme description; the mapping module is used for expanding the power field word family according to the service label to obtain the power field word family with service directivity; the retrieval module is used for performing retrieval according to the supplemented retrieval statement to obtain an initial retrieval result; and the sorting module is used for determining a retrieval result according to the sorting sequence. According to the method, the pre-training language model and the clustering algorithm are used for mining and semantic understanding of the power field knowledge, intelligent retrieval of the power field knowledge and effective matching of service information are achieved, and the stability and safety of the power dispatching system are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of power system operation and dispatching, and in particular to a control retrieval system and method based on word family recommendation. Background Art

[0002] In the field of power grid systems, as the scale of power grids gradually expands, the structure of power grids becomes increasingly complex, and the huge power grids put forward higher requirements on the power supply quality of power grids. As the core business of power dispatching, the amount of business data of dispatching management is also increasing, and a large amount of data generated in the process of power dispatching has been accumulated. The massive data contains rich value of power dispatching business, but there is also the problem of how to develop and utilize these data. Search engines have emerged to meet this demand. Search engines can quickly collect and retrieve information according to certain strategies and algorithms, provide retrieval services to users, and display relevant information to users.

[0003] At present, a large amount of data has been accumulated in the field of power grid dispatching. The data sources are diverse and the data relationships are complex. In the process of compilation, implementation, search, testing, and verification, there is a phenomenon that standard clauses do not match business processes. After the annotated text data is disassembled, there is a lack of accurate association with business information, resulting in low efficiency in searching the annotated text data. At the same time, business scenarios involve a variety of systems and different roles. It is difficult for fine-grained standard knowledge to efficiently associate system status data and role process information. Business personnel are faced with problems such as low efficiency in manually searching for regulations and operating processes, lack of efficient means of dispatching information retrieval, and difficulty in quickly grasping the overall information of power grid operation, which increases the decision-making time for business personnel to handle accidents.

[0004] In order to avoid low retrieval efficiency and inaccurate query results caused by mismatch between standard terms and business processes and insufficient data correlation during the retrieval of power dispatching business data, technical personnel in this field have been seeking a method for recommendation retrieval based on regulated word families, so as to achieve accurate matching of standard terms and business processes, improve the correlation between data and retrieval accuracy, and meet the personalized retrieval needs of users with different roles. Summary of the invention

[0005] The purpose of the present invention is to provide a control retrieval system and method based on word family recommendation, which utilizes pre-trained language models and clustering algorithms to perform knowledge mining and semantic understanding in the power field, and by constructing power field word families and performing business-oriented expansion, obtains power field knowledge and business-related information, realizes intelligent retrieval of power field knowledge and effective matching of business information, improves the level of intelligent information retrieval, reduces the workload of dispatchers, and enhances the stability and security of the power dispatching system.

[0006] The control retrieval system based on word family recommendation designed by the present invention to achieve one of the above purposes is special in that it includes:

[0007] An acquisition module, used for acquiring text data in the electric power field, and preprocessing the acquired text data in the electric power field to obtain preprocessed text data in the electric power field;

[0008] A construction module is used to segment the preprocessed text data in the electric power field through a segmentation model to obtain a word library in the electric power field; classify the word library in the electric power field based on word frequency to obtain a professional word library in the electric power field;

[0009] A parsing module is used to perform vector conversion on the professional vocabulary in the electric power field through a pre-trained language model to obtain word vectors of the professional vocabulary in the electric power field; perform cluster analysis on the word vectors of the professional vocabulary in the electric power field through a clustering algorithm to obtain word clusters of the professional vocabulary in the electric power field; perform topic extraction on the word clusters of the professional vocabulary in the electric power field to obtain topic descriptions corresponding to the word clusters of the professional vocabulary in the electric power field, and construct a word family in the electric power field according to the topic description;

[0010] A mapping module is used to formulate business labels according to business scenario requirements, perform vector conversion on the business labels through a pre-trained language model to obtain word vectors of the business labels, calculate the similarity between the word vectors of the business labels and the word vectors of the professional vocabulary in the electric power field, sort the word vector representations of the business labels using the similarity calculation results, and filter the word vectors of each sorted business label based on the similarity calculation results to obtain the business labels corresponding to the word vectors of the professional vocabulary in the electric power field, expand the electric power field word family according to the business labels, and obtain the electric power field word family with business orientation;

[0011] The retrieval module is used to obtain an input search statement, segment the search statement through a word segmentation model, search for a corresponding word family in the power field word family according to the segmented search statement, supplement the search statement through the corresponding word family, search according to the supplemented search statement, and obtain an initial search result.

[0012] Furthermore, the above system also includes: a sorting module; the sorting module is used to sort the initial search results according to a preset sorting rule, and determine the search results according to the arrangement order.

[0013] Furthermore, the construction module classifies the electric power field vocabulary based on word frequency, including: obtaining word frequency data of each word in the electric power field vocabulary in professional documents in the electric power field and in general documents in the electric power field, comparing the word frequency data of each word in the electric power field vocabulary with a preset word frequency threshold, and when the word frequency in the professional documents in the electric power field is greater than the professional word frequency threshold, and the word frequency in the general documents in the electric power field is less than the general word frequency threshold, the word is added to the professional word library in the electric power field.

[0014] Furthermore, the parsing module performs vector conversion on the professional vocabulary in the electric power field through a pre-trained language model, including: defining the professional vocabulary in the electric power field; assuming that the word set in the professional vocabulary in the electric power field is W = {w1, w2, ..., w i}; where w i Represents the i-th word in the word set; uses the multi-layer Transformer encoding layer of the pre-trained language model to encode the word w in the word set W i Encoding is performed, and the output set of the encoding layer is G = {h (1) ,h (2) ,…,h (L)}; where h (L) represents the output of the Lth encoding layer in the output set, where L represents the number of encoding layers of the pre-trained language model; calculate the word w i When the word vector is obtained, the output h of the last encoding layer is (L) As a word i The contextual semantic representation of is as follows: i )=h (L) ; The word vector of a word is weighted and fused with the semantic representation of the word to obtain a context-aware word vector, thereby enhancing the expressive power of the word vector.

[0015] Furthermore, the parsing module performs cluster analysis on the word vectors of the professional vocabulary in the electric power field by using a clustering algorithm, including: performing cluster calculation using the word vectors of each word, as described in the following formula:

[0016] N ε (v i )={v j ∈R d |||v i -v j ||≤ε}

[0017] Among them, N ε represents the neighborhood of a word, v i The word vector representing the word, v j Represents the word vector of the neighbor word, R d represents a vector space, ||vi -v j || represents the Euclidean distance between word vectors, and ε represents the radius of the neighborhood. Based on clustering calculation, multiple semantically similar word clusters are obtained, and each word cluster contains multiple words with similar semantic features in the power field.

[0018] Furthermore, the mapping module calculates the similarity between the word vector of the business tag and the word vector of the professional vocabulary in the power field, as described in the following formula:

[0019]

[0020] Among them, similarity(A,B) represents the similarity between the word vector of the business label and the word vector of the professional vocabulary in the power field, A represents the word vector of the business label; B represents the word vector of the professional vocabulary in the power field, ‖A‖ represents the modulus length of the word vector of the business label, and ‖B‖ represents the modulus length of the word vector of the professional vocabulary in the power field.

[0021] Furthermore, the retrieval module searches for a corresponding word family in the electric power field word family according to the search statement after word segmentation, including: grading based on the characteristics and logical relationships of the electric power field word family, indexing and weighting the electric power field word families at different levels; matching the search statement after word segmentation with the words in the electric power field word family based on the index and weight, and performing extended matching using synonyms, near-synonyms and hypernyms in the electric power field word family to obtain initial search results.

[0022] Furthermore, the sorting module sorts the initial search results according to preset sorting rules, including: sorting the initial search results in the following order: standard item name > standard chapter title > standard chapter text; if the initial search results contain repeated standard item names, sorting them in the following order: national standards > enterprise standards > industry standards; if the initial search results only involve the standard chapter text, sorting them in reverse order of word frequency.

[0023] The regulation retrieval method based on word family recommendation designed by the present invention to achieve the second objective above is special in that it comprises the following steps:

[0024] Acquire text data in the electric power field, and preprocess the acquired text data in the electric power field to obtain preprocessed text data in the electric power field;

[0025] The preprocessed text data in the electric power field is segmented by a word segmentation model to obtain a word library in the electric power field; the electric power field word library is classified based on word frequency to obtain a professional word library in the electric power field;

[0026] Performing vector conversion on the professional vocabulary in the electric power field through a pre-trained language model to obtain word vectors of the professional vocabulary in the electric power field; performing cluster analysis on the word vectors of the professional vocabulary in the electric power field through a clustering algorithm to obtain word clusters of the professional vocabulary in the electric power field; performing topic extraction on the word clusters of the professional vocabulary in the electric power field to obtain topic descriptions corresponding to the word clusters of the professional vocabulary in the electric power field, and constructing electric power field word families according to the topic descriptions;

[0027] Formulate business labels according to business scenario requirements, convert the business labels into vectors through a pre-trained language model to obtain word vectors of the business labels, calculate the similarity between the word vectors of the business labels and the word vectors of the professional vocabulary in the electric power field, sort the word vector representations of the business labels using the similarity calculation results, and filter the word vectors of each sorted business label based on the similarity calculation results to obtain the business labels corresponding to the word vectors of the professional vocabulary in the electric power field, expand the electric power field word family according to the business labels, and obtain a business-oriented electric power field word family;

[0028] The input search statement is obtained, the search statement is segmented by a word segmentation model, a corresponding word family is searched in the electric power field word family according to the segmented search statement, the search statement is supplemented by the corresponding word family, and a search is performed according to the supplemented search statement to obtain an initial search result.

[0029] An electronic device designed by the present invention to achieve the third objective above includes:

[0030] at least one processor; and

[0031] a memory communicatively connected to the at least one processor; wherein,

[0032] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in the present disclosure.

[0033] The present invention has the following beneficial effects:

[0034] (1) The control retrieval system and method based on word family recommendation converts word vectors through pre-trained language models, and uses clustering algorithms to analyze the semantic similarity of word vectors to construct professional word clusters in the power field. This semantic-based word family construction method overcomes the problem that synonyms and near-synonyms cannot be effectively matched in traditional keyword retrieval, and improves the coverage and recall rate of retrieval; based on the word family in the power field, the search sentence input by the user is supplemented, and the synonyms, near-synonyms and hypernyms in the word family are used for extended matching, which avoids the problem of incomplete or inaccurate search sentences caused by insufficient professional knowledge of users, and realizes intelligent retrieval.

[0035] (2) The control retrieval system and method based on word family recommendation optimizes the search matching efficiency by classifying the word families in the power field and indexing and weighting the word families at different levels. At the same time, the expansion of the search statement using the word families in the power field also reduces the number of search keywords that users need to enter; the intelligent search method based on word family recommendation reduces the need for manual intervention or modification of the search statement. Users can obtain relatively accurate search results without being proficient in professional knowledge, which reduces the usage threshold and improves the search efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A module schematic diagram showing a specific embodiment of a regulation retrieval system based on word family recommendation of the present invention.

[0037] Figure 2 A flowchart diagram showing a specific embodiment of a regulation retrieval method based on word family recommendation of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] like Figure 1 to Figure 2 As shown, an embodiment of the present invention discloses a control retrieval system and method based on word family recommendation, which utilizes the characteristics of text data generated in the historical dispatch retrieval process of the power grid, and obtains scene semantic keywords through text extraction and data analysis, thereby improving the level of intelligent information retrieval, reducing the workload of dispatchers, and enhancing the stability and security of the power dispatch automation system.

[0040] Example 1

[0041] This embodiment discloses a control retrieval system based on word family recommendation, the system comprising:

[0042] An acquisition module, used for acquiring text data in the electric power field, and preprocessing the acquired text data in the electric power field to obtain preprocessed text data in the electric power field;

[0043] A construction module is used to segment the preprocessed text data in the electric power field through a segmentation model. In specific implementation, Jieba segmentation can be used to segment the text data; obtain a word library in the electric power field; classify the word library in the electric power field based on word frequency to obtain a professional word library in the electric power field;

[0044] A parsing module is used to perform vector conversion on the professional vocabulary in the power field through a pre-trained language model. In specific implementation, the BERT (Bidirectional Encoder Representations from Transformers) Chinese pre-trained model can be used to vectorize the professional vocabulary in the power field; obtain the word vectors of the professional vocabulary in the power field; perform clustering analysis on the word vectors of the professional vocabulary in the power field through a clustering algorithm. In specific implementation, the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm can be used to cluster the word vectors; obtain the word clusters of the professional vocabulary in the power field, perform topic extraction on the word clusters of the professional vocabulary in the power field, and in specific implementation, the LDA (Latent Dirichlet Allocation) model can be used to generate clustering topics; obtain the topic description corresponding to the word clusters of the professional vocabulary in the power field, and construct the power field word family according to the topic description;

[0045] A mapping module is used to formulate business labels according to business scenario requirements, for example, "network protection" and "stable computing", and to convert the business labels into vectors through a pre-trained language model. In specific implementation, the pre-processed text training Word2Vec model can be used to vectorize the business labels; obtain the word vector of the business label, calculate the similarity between the word vector of the business label and the word vector of the professional vocabulary in the power field, sort the word vector representation of the business label using the similarity calculation result, and filter the word vectors of each sorted business label based on the similarity calculation result to obtain the business label corresponding to the word vector of the professional vocabulary in the power field, expand the power field word family according to the business label, and obtain the power field word family with business orientation;

[0046] The retrieval module is used to obtain an input search statement, segment the search statement through a word segmentation model, and search for a corresponding word family in the power field word family based on the segmented search statement. In specific implementation, an Elasticsearch distributed search engine can be used for retrieval; the search statement is supplemented by a corresponding word family, and a search is performed based on the supplemented search statement to obtain an initial search result.

[0047] Based on the above system, optionally, the system further includes: a sorting module; the sorting module is used to sort the initial search results according to a preset sorting rule, and determine the search results according to the arrangement order.

[0048] Specifically, the acquisition module acquires text data in the electric power field, including: terms or labels in electric power technical dictionaries, electric power scientific and technological papers, project reports, electric power regulations or electric power operation manuals; the acquisition module preprocesses the acquired text data in the electric power field, including: text cleaning, denoising, stop word removal and format unification processing of the text data.

[0049] Specifically, the construction module classifies the electric power field vocabulary based on word frequency, including: obtaining the word frequency data of each word in the electric power field vocabulary in professional documents in the electric power field and in general documents in the electric power field, comparing the word frequency data of each word in the electric power field vocabulary with a preset word frequency threshold, and when the word frequency of the word in the professional document in the electric power field is greater than the professional word frequency threshold, and the word frequency in the general document in the electric power field is less than the general word frequency threshold, the word is added to the professional word library in the electric power field.

[0050] Specifically, the parsing module performs vector conversion on the professional vocabulary in the electric power field through a pre-trained language model, including: defining the professional vocabulary in the electric power field; assuming that the word set in the professional vocabulary in the electric power field is W = {w1, w2, ..., w i}; where w i Represents the i-th word in the word set; uses the multi-layer Transformer encoding layer of the pre-trained language model to encode the word w in the word set W i Encoding is performed, and the output set of the encoding layer is H = {h (1) ,h (2) ,…,h (L)}; where h (L) represents the output of the Lth encoding layer in the output set, where L represents the number of encoding layers of the pre-trained language model; calculate the word w i When the word vector is obtained, the output h of the last encoding layer is (L) As a word i The contextual semantic representation of is as follows: i )=h (L) ; The word vector of a word is weighted and fused with the semantic representation of the word to obtain a context-aware word vector, thereby enhancing the expressive power of the word vector.

[0051] Specifically, the parsing module performs cluster analysis on the word vectors of the professional vocabulary in the electric power field through a clustering algorithm, including: performing cluster calculation using the word vectors of each word, as described in the following formula:

[0052] N ε (v i )={v j ∈R d |||v i -v j ||≤ε}

[0053] Among them, N ε represents the neighborhood of a word, v i The word vector representing the word, v j Represents the word vector of the neighbor word, R d represents a vector space, ||v i -v j | represents the Euclidean distance between word vectors, and ε represents the radius of the neighborhood. Based on clustering calculation, multiple semantically similar word clusters are obtained. The clustering result is C = {C1, C2, …, C m}, where m is the number of clusters, C m is a combination of several words w i The clusters formed by the word clusters are each composed of multiple words with similar semantic features in the power field. For example, if the high-frequency words of a cluster set are "power grid", "load", and "dispatching", then the topic description "power grid load dispatching" can be generated.

[0054] Specifically, the mapping module calculates the similarity between the word vector of the business label and the word vector of the professional vocabulary in the power field, as described in the following formula:

[0055]

[0056] Among them, similarity(A,B) represents the similarity between the word vector of the business label and the word vector of the professional vocabulary in the power field, A represents the word vector of the business label; B represents the word vector of the professional vocabulary in the power field, ‖A‖ represents the modulus length of the word vector of the business label, and ‖B‖ represents the modulus length of the word vector of the professional vocabulary in the power field.

[0057] Specifically, the retrieval module searches for the corresponding word family in the power field word family according to the search statement after word segmentation, including: grading based on the characteristics and logical relationships of the power field word family, indexing and weighting of different levels of the power field word family, and optionally including: separation of hot and cold data, optimizing data management and resource utilization through different storage strategies; matching the search statement after word segmentation with the words in the power field word family based on indexes and weights, and using synonyms, near synonyms and hypernyms in the power field word family for extended matching, for example, the power Synonyms include "electric energy" and "electric energy", which are similar in meaning to "electricity" and are often used to describe concepts related to electric energy. The initial search results are obtained by taking the "grid-related test" word family as an example. Its core words are "modeling test", "grid-related protection", "grid-connected performance test", etc., and related words are "grid-source coordination", "mechanical damping", "distributed power supply", etc. When searching through "grid-related test", based on the different levels and weights of various word segments in the "grid-related test" word family, similar files containing words such as "modeling test" and "grid-source coordination" are automatically indexed.

[0058] Specifically, the sorting module sorts the initial search results according to preset sorting rules, including: sorting the initial search results in the following order: standard item name > standard chapter title > standard chapter text; if the initial search results contain repeated standard item names, they are sorted in the following order: national standards > enterprise standards > industry standards; if the standard item name contains the words "distribution network", "power distribution", or "microgrid", it should be placed after other standard item names; if the initial search results only involve the standard chapter text, they are sorted in reverse order of word frequency.

[0059] Example 2

[0060] This embodiment discloses a control retrieval method based on word family recommendation, which includes the following steps:

[0061] Step 1: Acquire text data in the electric power field, and preprocess the acquired text data in the electric power field to obtain preprocessed text data in the electric power field;

[0062] Step 2: Segment the preprocessed text data in the electric power field through a word segmentation model to obtain a word library in the electric power field; classify the word library in the electric power field based on word frequency to obtain a professional word library in the electric power field;

[0063] Step 3: vectorize the professional vocabulary in the electric power field through a pre-trained language model to obtain word vectors of the professional vocabulary in the electric power field; cluster the word vectors of the professional vocabulary in the electric power field through a clustering algorithm to obtain word clusters of the professional vocabulary in the electric power field; perform topic extraction on the word clusters of the professional vocabulary in the electric power field to obtain topic descriptions corresponding to the word clusters of the professional vocabulary in the electric power field, and construct a word family in the electric power field according to the topic description;

[0064] Step 4: formulate business labels according to business scenario requirements, perform vector conversion on the business labels through a pre-trained language model to obtain word vectors of the business labels, calculate the similarity between the word vectors of the business labels and the word vectors of the professional vocabulary in the electric power field, sort the word vector representations of the business labels using the similarity calculation results, and screen the word vectors of each sorted business label based on the similarity calculation results to obtain the business labels corresponding to the word vectors of the professional vocabulary in the electric power field, expand the electric power field word family according to the business labels, and obtain the electric power field word family with business orientation;

[0065] Step 5, obtaining an input search statement, segmenting the search statement through a word segmentation model, searching for a corresponding word family in the electric power field word family according to the segmented search statement, supplementing the search statement through the corresponding word family, searching according to the supplemented search statement, and obtaining an initial search result;

[0066] Step 6: Sort the initial search results according to a preset sorting rule, and determine the search results according to the sorting order.

[0067] Example 3

[0068] This embodiment discloses an electronic device, and the specific embodiments of the present invention do not limit the specific implementation of the electronic device.

[0069] The electronic device includes a computing unit, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for the operation of the electronic device can also be stored. The computing unit, ROM and RAM are connected to each other through a bus. An input / output (I / O) interface is also connected to the bus.

[0070] Multiple components in an electronic device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks.

[0071] The computing unit may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of computing units include, but are not limited to, central processing units (CPUs), graphics processing units (GPUs), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit performs the various methods and processes described above, such as data processing methods. For example, in some embodiments, the data processing method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on an electronic device via a ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the computing unit, one or more steps of the data processing method described above may be performed. Alternatively, in other embodiments, the computing unit may be configured to perform the data processing method in any other appropriate manner (e.g., by means of firmware).

[0072] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0073] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, as an independent software package, partially on the machine and partially on a remote machine or completely on a remote machine or server. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described equipment and modules can refer to the corresponding process description in the aforementioned method embodiment, and will not be repeated here.

[0074] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the description of the above specific languages ​​is for disclosing the best mode of the present invention.

[0075] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0076] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the intention that the claimed invention requires more features than those expressly recited in each claim. Rather, as reflected in the claims, inventive aspects lie in less than all of the features of the individual embodiments previously disclosed. Therefore, the claims that follow the detailed description are hereby expressly incorporated into the detailed description, with each claim itself serving as a separate embodiment of the present invention.

Claims

1. A control retrieval system based on word family recommendation, characterized in that: include: An acquisition module, used for acquiring text data in the electric power field, and preprocessing the acquired text data in the electric power field to obtain preprocessed text data in the electric power field; A construction module is used to segment the preprocessed text data in the electric power field through a segmentation model to obtain an electric power field vocabulary; Classifying the electric power field vocabulary based on word frequency to obtain a professional vocabulary in the electric power field; A parsing module, used to perform vector conversion on the professional vocabulary in the electric power field through a pre-trained language model to obtain a word vector of the professional vocabulary in the electric power field; Performing cluster analysis on the word vectors of the professional vocabulary in the electric power field by using a clustering algorithm to obtain word clusters of the professional vocabulary in the electric power field, performing topic extraction on the word clusters of the professional vocabulary in the electric power field to obtain topic descriptions corresponding to the word clusters of the professional vocabulary in the electric power field, and constructing electric power field word families according to the topic descriptions; A mapping module is used to formulate business labels according to business scenario requirements, perform vector conversion on the business labels through a pre-trained language model to obtain word vectors of the business labels, calculate the similarity between the word vectors of the business labels and the word vectors of the professional vocabulary in the electric power field, sort the word vector representations of the business labels using the similarity calculation results, and filter the word vectors of each sorted business label based on the similarity calculation results to obtain the business labels corresponding to the word vectors of the professional vocabulary in the electric power field, expand the electric power field word family according to the business labels, and obtain the electric power field word family with business orientation; The retrieval module is used to obtain an input search statement, segment the search statement through a word segmentation model, search for a corresponding word family in the power field word family according to the segmented search statement, supplement the search statement through the corresponding word family, search according to the supplemented search statement, and obtain an initial search result.

2. The control retrieval system based on word family recommendation according to claim 1, characterized in that: Also includes: Sorting module: The sorting module is used to sort the initial search results according to a preset sorting rule and determine the search results according to the arrangement order.

3. The control retrieval system based on word family recommendation according to claim 1, characterized in that: The construction module classifies the electric power field vocabulary based on word frequency, including: obtaining word frequency data of each word in the electric power field vocabulary in professional documents and general documents in the electric power field, comparing the word frequency data of each word in the electric power field vocabulary with a preset word frequency threshold, and when the word frequency of the word in the professional document of the electric power field is greater than the professional word frequency threshold, and the word frequency in the general document of the electric power field is less than the general word frequency threshold, the word is added to the professional word library in the electric power field.

4. The control retrieval system based on word family recommendation according to claim 1, characterized in that: The parsing module performs vector conversion on the professional vocabulary in the electric power field through a pre-trained language model, including: defining the professional vocabulary in the electric power field; assuming that the word set in the professional vocabulary in the electric power field is W={w1, w2, ..., w i }; where w i Represents the i-th word in the word set; uses the multi-layer Transformer encoding layer of the pre-trained language model to encode the word w in the word set W i Encoding is performed, and the output set of the encoding layer is H = {h (1) ,h (2) , ..., h (L) }; where h (L) represents the output of the Lth encoding layer in the output set, where L represents the number of encoding layers of the pre-trained language model; calculate the word w i When the word vector is obtained, the output h of the last encoding layer is (L) As a word i The contextual semantic representation of is as follows: i )=h (L) ; The word vector of a word is weighted and fused with the semantic representation of the word to obtain a context-aware word vector, thereby enhancing the expressive power of the word vector.

5. The control retrieval system based on word family recommendation according to claim 1, characterized in that: The analysis module performs cluster analysis on the word vectors of the professional vocabulary in the electric power field by using a clustering algorithm, including: performing cluster calculation using the word vectors of each word, as described in the following formula: N ε (v i )={v j ∈R d |||v i -v j ||≤ε} Among them, N ε represents the neighborhood of a word, v i The word vector representing the word, v j Represents the word vector of the neighbor word, R d represents a vector space, ||v i -v j || represents the Euclidean distance between word vectors, and ε represents the radius of the neighborhood. Based on clustering calculation, multiple semantically similar word clusters are obtained, and each word cluster contains multiple words with similar semantic features in the power field.

6. The control retrieval system based on word family recommendation according to claim 1, characterized in that: The mapping module calculates the similarity between the word vector of the business label and the word vector of the professional vocabulary in the power field, as described in the following formula: Among them, similarity(A, B) represents the similarity between the word vector of the business label and the word vector of the professional vocabulary in the power field, A represents the word vector of the business label; B represents the word vector of the professional vocabulary in the power field, ||A|| represents the modulus of the word vector of the business label, and ||B|| represents the modulus of the word vector of the professional vocabulary in the power field.

7. The control retrieval system based on word family recommendation according to claim 1, characterized in that: The retrieval module searches for a corresponding word family in the electric power field word family according to the search statement after word segmentation, including: grading based on the characteristics and logical relationships of the electric power field word family, indexing and weighting the electric power field word families at different levels; matching the search statement after word segmentation with the words in the electric power field word family based on the index and weight, and performing extended matching using synonyms, near synonyms and hypernyms in the electric power field word family to obtain initial search results.

8. The control retrieval system based on word family recommendation according to claim 2, characterized in that: The sorting module sorts the initial search results according to preset sorting rules, including: sorting the initial search results in the following order: standard item name > standard chapter title > standard chapter text; if the initial search results contain repeated standard item names, sorting them in the following order: national standards > enterprise standards > industry standards; if the initial search results only involve the standard chapter text, sorting them in reverse order of word frequency.

9. A control retrieval method based on word family recommendation, characterized in that: The steps include: Acquire text data in the electric power field, and preprocess the acquired text data in the electric power field to obtain preprocessed text data in the electric power field; Segmenting the preprocessed text data in the electric power field through a word segmentation model to obtain an electric power field vocabulary; Classifying the electric power field vocabulary based on word frequency to obtain a professional vocabulary in the electric power field; Performing vector conversion on the professional vocabulary in the electric power field through a pre-trained language model to obtain a word vector of the professional vocabulary in the electric power field; Performing cluster analysis on the word vectors of the professional vocabulary in the electric power field by using a clustering algorithm to obtain word clusters of the professional vocabulary in the electric power field, performing topic extraction on the word clusters of the professional vocabulary in the electric power field to obtain topic descriptions corresponding to the word clusters of the professional vocabulary in the electric power field, and constructing electric power field word families according to the topic descriptions; Formulate business labels according to business scenario requirements, convert the business labels into vectors through a pre-trained language model to obtain word vectors of the business labels, calculate the similarity between the word vectors of the business labels and the word vectors of the professional vocabulary in the electric power field, sort the word vector representations of the business labels using the similarity calculation results, and filter the word vectors of each sorted business label based on the similarity calculation results to obtain the business labels corresponding to the word vectors of the professional vocabulary in the electric power field, expand the electric power field word family according to the business labels, and obtain a business-oriented electric power field word family; The input search statement is obtained, the search statement is segmented by a word segmentation model, a corresponding word family is searched in the electric power field word family according to the segmented search statement, the search statement is supplemented by the corresponding word family, and a search is performed according to the supplemented search statement to obtain an initial search result.

10. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the regulation retrieval method based on word family recommendation described in claim 9.

Images