A method for constructing a knowledge base of highway electromechanical operation and maintenance based on big data

Abstract

The present invention provides a method for constructing a highway electromechanical operation and maintenance knowledge base based on big data, the method comprising: (1) collecting highway electromechanical operation and maintenance text data; (2) preprocessing the highway electromechanical operation and maintenance text data; (3) training a weighted LDA model with the processed data to perform topic mining and classification on the text base; (4) constructing a highway electromechanical operation and maintenance knowledge base; (5) optimizing the highway electromechanical operation and maintenance knowledge base. The present invention provides a method for constructing a highway operation and maintenance knowledge base by using an improved LDA model, and ensures the systematicness and accuracy of the constructed knowledge base; and optimizes the knowledge base by using data indexing, classification algorithms and knowledge recommendation, thereby further improving query efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital electromechanical operation and maintenance management of highways, and specifically relates to a method for constructing a highway electromechanical operation and maintenance knowledge base based on big data. Background Art

[0002] With the continuous expansion of highway network and the application of new technologies, the complexity and professionalism of electromechanical operation and maintenance management have become increasingly prominent, which is reflected in many aspects such as a wide variety of equipment, rapid technology updates, and strict maintenance requirements. At present, a large amount of internal and external text data resources have been accumulated in the field of highway electromechanical operation and maintenance, but due to different data sources, data structures, and data standards, its storage has problems such as data redundancy, dispersion, and low relevance, which makes it difficult for operation and maintenance personnel to quickly obtain key information and effective knowledge from the data.

[0003] Building a comprehensive, systematic and efficient electromechanical operation and maintenance knowledge base can not only improve the technical level and emergency response capability of the operation and maintenance team, but also effectively ensure the stable operation of the electromechanical system of the highway and improve the operational efficiency and service quality of the entire transportation network. In addition, the establishment of a knowledge base can also promote information sharing and knowledge inheritance, and provide solid knowledge support for the sustainable development of electromechanical operation and maintenance of highways. Therefore, building a knowledge base for electromechanical operation and maintenance of highways is a necessary measure to improve the level of operation and maintenance management, and it has far-reaching significance for ensuring the safe, smooth and efficient operation of highways. Summary of the invention

[0004] The present invention provides a method for constructing a highway electromechanical operation and maintenance knowledge base based on big data, wherein the method comprises

[0005] (1) Collecting highway electromechanical operation and maintenance text data;

[0006] (2) Preprocessing of highway electromechanical operation and maintenance text data;

[0007] (3) The processed data is used to train a weighted LDA model to perform topic mining and classification on the text library;

[0008] (4) Build a knowledge base for highway electromechanical operation and maintenance;

[0009] (5) Optimize the knowledge base of highway electromechanical operation and maintenance.

[0010] Furthermore, the step (1) includes collecting text data of highway operation and maintenance including highway electromechanical operation and maintenance knowledge documents, business analysis reports, operation and maintenance service work orders, industry standards and policies, product manuals and meeting minutes, and aggregating them into a highway electromechanical operation and maintenance text library.

[0011] Furthermore, the step (2) includes cleaning the collected text data, performing word segmentation and part-of-speech tagging on the text data, splitting the text into independent words, filtering out stop words and useless words, and performing word frequency statistics on each processed text.

[0012] Furthermore, the step (3) comprises

[0013] (3.1) Generate the vocabulary vector matrix Corpus based on the bag of words and corpus;

[0014] (3.2) Calculate the weight values ​​and convert the Corpus matrix into an integer vocabulary vector matrix Corpus_Weighted_Int;

[0015] (3.3) Initialize model parameters, determine the number of topics K, and set hyperparameters ɑ and β;

[0016] (3.4) For each word in each document, randomly assign a topic and update the topic using Gibbs sampling;

[0017] (3.5) Repeat step (3.4) until Gibbs sampling converges or reaches the preset number of iterations;

[0018] (3.6) Count the final topic-word co-occurrence frequency matrix, which is the trained LDA model;

[0019] (3.7) Use the P value to semantically expand the data after topic mining by LDA model.

[0020] Furthermore, the step (3.6) includes that the LDA model comprises a three-layer structure of vocabulary, topic and document, let w represent vocabulary, z represent topic, K represent the total number of topics, M represent the total number of documents, N represent the total number of vocabulary of documents, θ is the Dirichlet distribution of document-topic, is the Dirichlet distribution of topic-words, ɑ and β are θ and Hyperparameters of

[0021] The joint division calculation formula of the LDA model is:

[0022] P(w,z|α,β)=∑P(w|z,β)P(z|α)

[0023] Use Gibbs sampling to fit θ and θ and The calculation formula is as follows:

[0024]

[0025]

[0026] in, is the number of words for topic j in document d, is the number of words for all topics in document d, is the frequency of word w in topic j, is the total number of words for topic j in document d.

[0027] Furthermore, the step (3.7) comprises

[0028] Construct a probability P value of an ordered phrase. The formula for the P value is as follows:

[0029]

[0030] Where word1 and word2 represent two different words, P(word1) represents the probability of word 1 appearing, and P(word2) represents the probability of word 2 appearing. word1word2 represents a phrase composed of words 1 and 2 connected in order from left to right, and P(word1word2) is the probability of the phrase appearing.

[0031] Arrange in descending order of P value, compare with the original subject words and then replace them to achieve the expansion of subject words by keyword blocks.

[0032] Furthermore, step (4) includes using the trained improved LDA model topic clusters as entries of the knowledge base, classifying the corpus and dividing it into different topics, sorting the topics and using the topic classification results as labels of the knowledge base, dividing each document into different topics, and storing the documents under each topic in the knowledge base to form a highway operation and maintenance knowledge base.

[0033] Furthermore, the step (5) includes optimizing the knowledge base by using data indexing and classification algorithms, and implementing knowledge recommendation based on document similarity to improve query efficiency and accuracy.

[0034] Furthermore, the step (5) comprises

[0035] Index documents in the knowledge base for users to quickly find and search;

[0036] Classify documents in the knowledge base by subject and type to organize and manage the knowledge base;

[0037] Regularly update the documents in the knowledge base, update the latest documents and remove outdated documents;

[0038] Realize knowledge recommendation in knowledge base based on document similarity.

[0039] Furthermore, knowledge recommendation based on document similarity includes

[0040] Use JS distance as the similarity metric. JS distance is obtained by transforming KL distance. The calculation method of KL distance is shown in the following formula:

[0041]

[0042] The calculation method of JS distance is shown in the following formula:

[0043]

[0044] The range of JS distance is [0,1]. When JS distance is close to 0, the similarity between two documents is higher. When JS distance is close to 1, the similarity between two documents is lower.

[0045] Sort the document by its JS distance from all other documents, and recommend documents from high to low based on similarity.

[0046] Beneficial effects:

[0047] The present invention provides a method for constructing a highway operation and maintenance knowledge base based on big data by using an improved LDA model, and ensures the systematicness, accuracy and interpretability of the constructed knowledge base; at the same time, the present invention realizes the efficient organization and management of the knowledge base through functions such as data indexing, classification algorithm and knowledge recommendation, further improves the product effect and user experience of the knowledge base, and can effectively promote the improvement of the electromechanical operation and maintenance level.

[0048] It should be understood that the above general description and the following specific embodiments are only exemplary and illustrative and do not limit the scope of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flowchart of a method for constructing a highway electromechanical operation and maintenance knowledge base based on big data according to the present invention;

[0050] Figure 2 It is a structural schematic diagram of an LDA model in a method for constructing a highway electromechanical operation and maintenance knowledge base based on big data according to the present invention;

[0051] It should be understood that the drawings are not necessarily drawn to scale, presenting somewhat simplified representations of various features illustrative of the basic principles of the present disclosure. The specific design features of the present invention as disclosed herein, including, for example, specific dimensions, orientations, locations, and shapes will be determined in part by the particular intended application and use environment.

[0052] In the drawings, reference numbers refer to the same or equivalent parts of the present invention throughout the several figures of the drawing. DETAILED DESCRIPTION

[0053] Reference will now be made in detail to various embodiments of the present invention, examples of which are illustrated in the accompanying drawings and described below. Although the present invention will be described in conjunction with exemplary embodiments of the present invention, it should be understood that this specification is not intended to limit the present invention to those exemplary embodiments. On the other hand, the present invention is intended to cover not only the exemplary embodiments of the present invention, but also various alternatives, modifiers, equivalents and other embodiments, which may be included in the spirit and scope of the present invention as defined by the appended claims.

[0054] Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. The specific structures and functions described in the exemplary embodiments of the present invention are for illustrative purposes only. Embodiments according to the concepts of the present invention can be implemented in various forms, and it should be understood that they should not be interpreted as being limited to the exemplary embodiments described in the exemplary embodiments, but include all modifications, equivalents or substitutes included in the spirit and scope of the present invention.

[0055] Throughout the specification, the professional terms used herein are only for describing various exemplary embodiments and are not intended to be limiting. It will be further understood that the terms "include", "comprising", "having", etc., when used in the exemplary embodiments, specifically refer to the presence of the stated parts, steps, operations or elements, but do not exclude the presence or addition of one or more other parts, steps, operations or elements.

[0056] The present invention aims to perform topic mining and classification on text data related to highway electromechanical operation and maintenance by improving the LDA model, so as to effectively integrate the massive data resources of highway electromechanical operation and maintenance, and build a systematic and scientific professional knowledge base in the field of highway electromechanical operation and maintenance to achieve knowledge sharing, rapid fault diagnosis and decision support.

[0057] like Figure 1 As shown, the present invention provides a method for constructing a highway electromechanical operation and maintenance knowledge base based on big data, the method comprising:

[0058] Step 1: Collect highway electromechanical operation and maintenance text data.

[0059] A large amount of highway electromechanical operation and maintenance text corpus data related to highway operation and maintenance is collected, including highway electromechanical operation and maintenance knowledge documents, business analysis reports, operation and maintenance service work orders, industry standards and policies, product manuals, meeting minutes, etc., and summarized into a highway electromechanical operation and maintenance text library.

[0060] The collected text data is automatically deduplicated based on semantic similarity matching, invalid data is removed, and the text library is corrected.

[0061] Step 2: Preprocess the highway electromechanical operation and maintenance text data.

[0062] Jieba is used to clean the collected text data, perform word segmentation and part-of-speech tagging on the text data, split the text into independent words, filter out stop words and useless words, and perform word frequency statistics on each processed text. Finally, a bag-of-words dictionary is formed to prepare data for LDA model analysis.

[0063] When using Jieba to segment text, by entering user-defined industry dictionaries, the Jieba segmentation process can be more adapted to professional vocabulary related to highway electromechanical operation and maintenance, thereby improving the accuracy and efficiency of segmentation.

[0064] Step 3: Use the processed data to train a weighted LDA model to perform topic mining and classification on the text library.

[0065] The processed data is used to train the weighted LDA model, and the text library is subjected to topic mining and classification. The weighted LDA model introduces a weight factor to perform differentiated processing on words of different properties to enhance the model's ability to capture operation and maintenance knowledge. In addition, the improved method of the weighted LDA model proposed in the present invention can enhance the LDA model's mining of word order and phrases, thereby improving the interpretability and recognizability of topic mining results.

[0066] The LDA (Latent Dirichlet Allocation) model is a document topic generation model. It is a three-layer Bayesian probability model that can automatically divide documents in a corpus into different topics. The LDA model contains a three-layer structure of words, topics, and documents. The model believes that a document contains several topics, and each topic is represented by several words. The schematic diagram of the LDA model is as follows: Figure 2 As shown. Where w represents vocabulary, z represents topic, K represents the total number of topics, M represents the total number of documents, N represents the total number of vocabulary in the document, and θ is the Dirichlet distribution of "document-topic". is the Dirichlet distribution of "topic-word", ɑ and β are θ and Hyperparameters of .

[0067] The joint division calculation formula of the LDA model is:

[0068] P(w,z|α,β)=∑P(w|z,β)P(z|α) (1)

[0069] Use Gibbs sampling to fit θ and θ and The calculation formula is as follows:

[0070]

[0071]

[0072] in, is the number of words for topic j in document d, is the number of words for all topics in document d, is the frequency of word w in topic j, is the total number of words for topic j in document d.

[0073] Furthermore, the present invention innovatively proposes a weight parameter calculation method, which comprehensively evaluates the contribution of each word to the document topic by comprehensively considering the frequency, diversity, and uniformity of the word distribution within the document. i The weight in document d is W i,d , then the weight of the word is determined by the following formula:

[0074]

[0075] Among them, TF(w i ,d) is the word frequency, which indicates the frequency of word i appearing in document d;

[0076] IDF(w i ) is the inverse document frequency, which indicates how rare word i is in the entire corpus;

[0077] UD(w i ,d) is the uniform distribution degree, which indicates the uniformity of the distribution of word i in each sentence or paragraph in document d. NF is used as a normalization factor to ensure that the sum of the weights of all words is 1.

[0078] Furthermore, the calculation formulas for term frequency (TF), inverse document frequency (IDF), uniform distribution (UD) and normalization factor (NF) are as follows.

[0079]

[0080] Among them, n wi,d It is the vocabulary w i The denominator is the sum of the number of times a word appears in document d.

[0081]

[0082] Where N is the total number of documents in the corpus, and is a word containing the word w i The total number of documents.

[0083]

[0084] Among them, var(positions of w i in d) is the vocabulary w i Variance of occurrence positions in document d.

[0085]

[0086] Among them, V d is the set of all distinct words in document d.

[0087] The above weight calculation method provides a comprehensive and balanced importance assessment for each word by comprehensively considering the frequency of the word, its distribution in the entire corpus, and its uniformity in the document. Normalization further ensures the interpretability of the weights and the stability of the model, making the results easier to understand and use. In this way, the model can more accurately identify and extract the key topics of the document, improving the performance of topic modeling and the quality of the results.

[0088] The weight calculation method of the present invention replaces the word frequency in the traditional LDA model algorithm to optimize the LDA model. The implementation process of the improved weighted LDA model is as follows:

[0089] (1) Generate word vector matrix Corpus: Create a two-dimensional matrix M×N based on the bag of words and corpus, where M is the number of documents and N is the size of the vocabulary. Fill the matrix, and each element represents the number of times the corresponding word appears in the corresponding document. For example, if a word does not appear in the document, then a ij =0, if a word appears 3 times in the document, then a ij =3.

[0090] (2) Calculate the weight value: For each non-zero element in the Corpus matrix, calculate its weight value W i,d . And transform it into an integer type, and obtain the integer vocabulary vector matrix Corpus_Weighted_Int.

[0091] (3) Initialize model parameters: determine the number of topics K and set hyperparameters ɑ and β.

[0092] (4) Topic assignment and update: For each word in each document, a topic is randomly selected. Then the Gibbs sampling algorithm is used to update its topic assignment in the corpus. During the update process, the probability of topic assignment is adjusted according to the weighted value of the word, so that more important words have a greater impact on the topic assignment.

[0093] (5) Convergence check and iteration: Repeat step (4) until Gibbs sampling converges or reaches the preset number of iterations.

[0094] (6) Model training is completed: The final topic-word co-occurrence frequency matrix is ​​calculated, which is the trained LDA model.

[0095] However, the above technical solutions still have room for improvement in the field of highway electromechanical operation and maintenance data analysis. For example, the words containing "facilities" may be monitoring facilities, communication facilities, toll facilities, power supply and distribution facilities, lighting facilities, tunnel electromechanical facilities, etc., and only counting the word "facilities" will greatly reduce the granularity of data analysis; for example, the words containing "license plate recognition" may be license plate recognition rate or license plate recognition device. The nature of the two is completely different, and should not be simply summarized by the word "license plate recognition". In the process of text mining, these phrases cannot be split into independent words for simple understanding, otherwise the original meaning of the text will be lost; therefore, word order and phrases are particularly critical for the content representation of text research in the field of highway electromechanical operation and maintenance.

[0096] Therefore, this patent innovatively proposes an improved method of the LDA model: by calculating the co-occurrence probability between two ordered words, the correlation between the words is obtained. This method constructs the probability of occurrence of an ordered phrase (P value). The larger the P value, the higher the certainty that the words contained in the phrase form a phrase. The formula for the P value is as follows:

[0097]

[0098] Among them, word1 and word2 represent two different words, P(word1) represents the probability of word 1 appearing, and P(word2) represents the probability of word 2 appearing. word1word2 represents a phrase composed of words 1 and 2 in order from left to right, and P(word1word2) is the probability of the phrase appearing. It should be noted that since there is a sequence in the combination of words, word1word2 and word2word1 are two different words, and their corresponding probabilities are not necessarily equal.

[0099] (7) Use the P value to semantically expand the data after topic mining by the LDA model.

[0100] Extract the common and important phrases with high co-occurrence frequency, arrange them in descending order according to the P value, and replace them after comparison with the original subject words to achieve the expansion of the subject words by keyword blocks, thereby achieving a significant improvement in semantic interpretability.

[0101] Step 4: Build a knowledge base for highway electromechanical operation and maintenance.

[0102] According to the trained improved LDA model topic clusters as entries of the knowledge base, the corpus is classified and divided into different topics. Referring to the topic strength calculated by the LDA model, the topics are sorted and the topic classification results are used as labels of the knowledge base. Each document is divided into different topics, and the documents under each topic are stored in the corresponding knowledge base to form a highway operation and maintenance knowledge base.

[0103] Step 5: Optimize the highway electromechanical operation and maintenance knowledge base.

[0104] The knowledge base is optimized using data indexing and classification algorithms, and related knowledge is recommended based on document similarity to improve query efficiency and accuracy. Specifically, it includes: 1. Indexing documents in the knowledge base so that users can quickly find and search; 2. Classifying documents in the knowledge base by subject and type to better organize and manage the knowledge base; 3. Regularly updating documents in the knowledge base, adding the latest documents and removing outdated documents; 4. Implementing the related knowledge recommendation function of the knowledge base based on document similarity.

[0105] In the above technical solution, the knowledge recommendation method based on document similarity is as follows:

[0106] Since the topic distribution of a document is a simplex mapping of the document vector space, in the case of a document's topic representation, the similarity between two documents can be calculated by calculating the corresponding topic probability distribution. Since the topic is a mixed distribution of word vectors, the JS distance is used as the similarity metric; the JS distance is obtained by transforming the KL distance, and the calculation method of the KL distance is shown in Formula 6:

[0107]

[0108] The calculation method of JS distance is shown in Formula 7:

[0109]

[0110] The range of JS distance is [0,1]. The closer the JS distance is to 0, the higher the similarity between the two documents. The closer the JS distance is to 1, the lower the similarity between the two documents. Therefore, by sorting the JS distance between a document and all other documents from small to large, related documents can be recommended from high to low in terms of similarity.

[0111] The foregoing description of specific exemplary embodiments of the present invention has been presented for the purpose of illustration and description. It is not intended to exclude or limit the present invention to the precise form disclosed, and it is obvious that many modifications and changes are feasible in view of the above teachings. Exemplary embodiments are selected and described to explain certain principles of the present invention and their practical application, so that other technicians in the field can make or use various exemplary embodiments of the present invention, and various alternatives and modifications thereof. It is intended that the scope of the present invention will be defined by the claims attached hereto and their equivalents.

[0112] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for constructing a highway electromechanical operation and maintenance knowledge base based on big data, characterized in that: The method comprises the following steps: (1) Collecting highway electromechanical operation and maintenance text data; (2) Preprocessing of highway electromechanical operation and maintenance text data; (3) The processed data is used to train a weighted LDA model to perform topic mining and classification on the text library; (3.1) Generate the vocabulary vector matrix Corpus based on the bag of words and corpus; (3.2) Calculate the weight values ​​and convert the Corpus matrix into an integer vocabulary vector matrix Corpus_Weighted_Int; Setting vocabulary The weight in document d is , the weight is determined by the following formula: ； in, is the word frequency, which indicates the frequency of word i appearing in document d; is the inverse document frequency, which indicates how rare word i is in the entire corpus; is the uniform distribution degree, which indicates the uniformity of the distribution of word i in each sentence or paragraph in document d; As a normalization factor, ensure that the sum of all vocabulary weights is 1; The calculation formulas for term frequency TF, inverse document frequency IDF, uniform distribution UD and normalization factor NF are as follows; ； in, It is vocabulary The number of times it appears in document d, and the denominator is the sum of the number of times all words appear in document d; ； in, is the total number of documents in the corpus, and Contains vocabulary Total number of documents; ； in, It is vocabulary The variance of the occurrence position in document d; ； in, is the set of all distinct words in document d; (3.3) Initialize model parameters, determine the number of topics K, and set hyperparameters ɑ and β; (3.4) For each word in each document, randomly assign a topic and update the topic using Gibbs sampling; (3.5) Repeat step (3.4) until Gibbs sampling converges or reaches the preset number of iterations; (3.6) Calculate the final topic-word co-occurrence frequency matrix, which is the trained LDA model; (3.7) Use the P value to semantically expand the data after the LDA model is used for topic mining; By calculating the co-occurrence probability between two ordered words, the correlation between the words is obtained; the occurrence probability P value of an ordered phrase is constructed. The larger the P value, the higher the certainty that the words contained in the phrase form a phrase. The formula of the P value is as follows: ； in, and Indicates two different words. represents the probability of word 1 appearing, Indicates the probability of word 2 appearing; It means that words 1 and 2 are connected from left to right to form a phrase. is the probability of the phrase appearing; Extract common and important phrases with high co-occurrence frequency, arrange them in descending order according to P value, and replace them after comparison with the original subject words, so as to expand the subject words with keyword blocks and achieve a significant improvement in semantic interpretability; (4) Build a knowledge base for highway electromechanical operation and maintenance; (5) Optimize the knowledge base of highway electromechanical operation and maintenance.

2. According to the method for constructing a highway electromechanical operation and maintenance knowledge base based on big data in claim 1, it is characterized in that: The step (1) includes collecting text data of highway operation and maintenance, including highway electromechanical operation and maintenance knowledge documents, business analysis reports, operation and maintenance service work orders, industry standards and policies, product manuals and meeting minutes, and aggregating them into a highway electromechanical operation and maintenance text library.

3. The method for constructing a highway electromechanical operation and maintenance knowledge base based on big data according to claim 1 is characterized in that: The step (2) includes cleaning the collected text data, performing word segmentation and part-of-speech tagging on the text data, splitting the text into independent words, filtering out stop words and useless words, and performing word frequency statistics on each processed text.

4. The method for constructing a highway electromechanical operation and maintenance knowledge base based on big data according to claim 1 is characterized in that: The step (3.6) includes that the LDA model includes a three-layer structure of vocabulary, topic and document, let w represent vocabulary, z represent topic, K represent the total number of topics, M represent the total number of documents, N represent the total number of vocabulary in the document, θ is the Dirichlet distribution of document-topic, is the Dirichlet distribution of topic-words, ɑ and β are θ and Hyperparameters of The joint division calculation formula of the LDA model is: ； Use Gibbs sampling to fit θ and , θ and The calculation formula is as follows: ； ； in, is the number of words for topic j in document d, is the number of words for all topics in document d, is the frequency of word w in topic j, is the total number of words for topic j in document d.

5. The method for constructing a highway electromechanical operation and maintenance knowledge base based on big data according to claim 1 is characterized in that: The step (4) includes taking the trained improved LDA model topic clusters as entries of the knowledge base, classifying the corpus and dividing it into different topics, sorting the topics and using the topic classification results as labels of the knowledge base, dividing each document into different topics, and storing the documents under each topic in the knowledge base to form a highway operation and maintenance knowledge base.

6. The method for constructing a highway electromechanical operation and maintenance knowledge base based on big data according to claim 1 is characterized in that: The step (5) includes optimizing the knowledge base by using data indexing and classification algorithms, and implementing knowledge recommendation based on document similarity to improve query efficiency and accuracy.

7. The method for constructing a highway electromechanical operation and maintenance knowledge base based on big data according to claim 6 is characterized in that: The step (5) includes Index documents in the knowledge base for users to quickly find and search; Classify documents in the knowledge base by subject and type to organize and manage the knowledge base; Regularly update the documents in the knowledge base, update the latest documents and remove outdated documents; Realize knowledge recommendation in knowledge base based on document similarity.

8. According to the method for constructing a highway electromechanical operation and maintenance knowledge base based on big data in claim 7, the knowledge recommendation based on document similarity includes: Use JS distance as the similarity metric. JS distance is obtained by transforming KL distance. The calculation method of KL distance is shown in the following formula: ； The calculation method of JS distance is shown in the following formula: ； in, The range of JS distance is [0,1]. When JS distance is close to 0, the similarity between two documents is higher; when JS distance is close to 1, the similarity between two documents is lower. Sort the document by its JS distance from all other documents, and recommend documents from high to low based on similarity.

Images