CN116433799A - A flow chart generation method and device based on semantic similarity and subgraph matching - Google Patents

Abstract

The invention discloses a flow chart generation method based on semantic similarity and subgraph matching, comprising: obtaining user requirement documents and RPA project asset library documents; calculating the semantic similarity between user requirement documents and RPA project asset library documents, and according to the semantic similarity Rank the subgraphs to be matched from high to low to obtain the top-k subgraphs to be matched; construct the query knowledge graph according to the user demand document, and set a start node; search and match the subgraphs to be matched in the query knowledge graph to obtain the final best Match the graph and return the corresponding flowchart. The technical scheme of the invention reduces the time for searching and traversing the subgraph matching, doubles constraints on semantics and structure, and more accurately generates the flow chart required by the current user.

Description

A Flowchart Generation Method and Device Based on Semantic Similarity and Subgraph Matching

technical field

The invention belongs to the technical field of data processing, and in particular relates to a flow chart generation method and device based on semantic similarity and subgraph matching.

Background technique

For the execution of business processes, employees currently spend significant time dealing with enterprise resource planning (ERP), customer relationship management (CRM), spreadsheets, and legacy systems, performing manual repetitive tasks such as entry, copy, paste, extract, merge, and move Large amounts of data are passed from one system to another. Considering that some of these highly structured, routine, and manual tasks can be handled by robots, this leaves knowledge workers more time for value-added tasks. Robotic Process Automation (RPA) emerged as a software-based solution for automating rule-based business processes involving routine tasks, structured data, and deterministic outcomes. The flow chart is an important part of RPA technology. The flow chart is drawn according to the user's needs, and then the execution code is generated according to the flow chart to complete the specified operation. Flowcharts are extremely helpful in understanding exactly how things are going and deciding how the process should be improved. However, manual drawing of flow charts takes a long time and consumes more human resources. How to use the existing flowchart in the RPA implementation library to automatically generate a flowchart for the current user needs can save manpower and material resources, and greatly improve the efficiency of RPA implementation, which is the direction of current research.

Contents of the invention

In view of this, the present invention proposes a flow chart generation method based on semantic similarity and subgraph matching, including the following steps:

S1. Obtain user requirement documents and RPA project asset library documents;

S2. Calculate the semantic similarity between the user requirement document and the RPA project asset library document, and arrange top-k subgraphs to be matched according to the semantic similarity from high to low;

S3. Construct a query knowledge map according to the user requirement document, and set a start node;

S4. Search and match the subgraphs to be matched in the query knowledge graph, obtain the final best matching graph, and return the corresponding flow chart.

The present invention also proposes a flow chart generation device based on semantic similarity and subgraph matching, including:

processor;

a memory on which is stored a computer program executable on said processor;

Wherein, when the computer program is executed by the processor, a flowchart generation method based on semantic similarity and subgraph matching is realized.

The beneficial effects brought by the technical scheme provided by the invention are:

The technical solution proposed by the present invention utilizes the user's demand document to filter out items similar to the current demand in the project library through semantic similarity, and at the same time constructs the user's demand document into a small demand graph and uses the method of fuzzy subgraph matching to match RPA project assets The knowledge map is matched to find items similar to the current user demand process structure. First use semantics for preliminary screening, and then use subgraph matching for search traversal. On the one hand, it reduces the time for searching and traversing subgraph matching, and on the other hand, it generates the flow chart of current user needs more accurately from the dual constraints of semantics and structure. Using generation technology to automatically generate flowcharts according to user needs, which is convenient for users to use directly or fine-tune, simplifies the process of drawing flowcharts, greatly reduces manual intervention, and improves the efficiency of RPA.

Description of drawings

Fig. 1 is a flow chart of a flow chart generation method based on semantic similarity and subgraph matching in an embodiment of the present invention;

Fig. 2 is the Agricultural Bank of China online banking running water download user demand knowledge map constructed by the embodiment of the present invention;

Fig. 3 is the knowledge map of online banking flow downloading of Company B's RPA asset library project constructed by the embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solution and advantages of the present invention clearer, the embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

A method for generating a flowchart based on semantic similarity and subgraph matching proposed by the present invention refers to FIG. 1. FIG. 1 is a flowchart of a method for generating a flowchart based on semantic similarity and subgraph matching in an embodiment of the present invention, including the following steps :

S1. Obtain user requirement documents and RPA project asset library documents.

In a further embodiment, the user's demand is the online banking flow download of the Agricultural Bank of China.

S2. Calculate the semantic similarity between the user requirement document and the RPA project asset library document, and arrange top-k subgraphs to be matched according to the semantic similarity from high to low.

Specifically:

进行分词，并去除停用词。S21. Requirements document Q for users and user documents of RPA project asset library

Perform word segmentation and remove stop words.

进行分词。In this embodiment, the word segmentation component jieba is used to analyze the user's requirement document Q and the RPA project asset library user document

Participate.

表示成一维的向量。S22. Use the frequency of words in the text to describe the document, and combine the user's demand document Q and the user document of the RPA project asset library

Represented as a one-dimensional vector.

。In this embodiment, the normalized BOW bag-of-words model is used to represent Q and

.

的每个词语的嵌入向量，在这个向量空间中，语义相似的词之间距离相近。S23. According to the one-dimensional vector obtained in S22, learn each word of the user requirement document Q and the user document of the RPA project asset library

The embedding vector of each word in , in this vector space, the distance between semantically similar words is similar.

的每个词语的嵌入向量。In this embodiment, the word2vec method is used to learn each word of the user demand document Q and the user document of the RPA project asset library

Embedding vectors for each word in .

之间的文本相似度，WMD算法将文档距离建模成两个文档中词的语义距离的组合，对两个文档中的任意两个词所对应的词向量求欧氏距离然后再加权求和。WMD算法是基于word2vec基础上通过计算文本间词的距离来衡量文本相似度的算法。S24, use the WMD algorithm to calculate the user requirement document Q and the user document of the RPA project asset library

The text similarity between the documents, the WMD algorithm models the document distance as a combination of the semantic distances of the words in the two documents, and calculates the Euclidean distance for the word vectors corresponding to any two words in the two documents, and then reweights the summation . The WMD algorithm is an algorithm based on word2vec to measure text similarity by calculating the distance between words in text.

中的词出现的次数进行归一化处理，计算文档Q中第i个词汇的词频/>

，文档/>

中第j个词汇的词频为/>

：S241, document Q and document

The number of occurrences of the words in the document Q is normalized to calculate the word frequency of the i-th vocabulary in the document Q/>

, document />

The word frequency of the jth vocabulary in is />

:

、/>

分别表示文档Q、文档D_s中第i个词、第j个词出现的次数；Among them, m and n are the number of words in document Q and document D _s respectively,

, />

Respectively represent the number of occurrences of the i-th word and j-th word in document Q and document D _s ;

：S242. Calculate the Euclidean distance between the word i from the document Q and the word j from the document D _s as

:

和/>

为词i和j学习到的嵌入向量；in,

and />

Embedding vectors learned for words i and j;

S243, using a dynamic programming algorithm to solve the document Q and the WMD distance of each document in the RPA project asset library D _s :

表示将文档Q中的单词i映射到文档D_s中的单词j的权重，/>

表示文档Q中的单词i和文档D_s中的单词j之间的距离；值得注意的是，文档Q的第i个单词对应到D_s的一个文档中所有单词的权重值的和等于/>

，同理，D_s中的文档的第j个单词映射到文档Q的所有单词的权重值的和等于/>

，其中，f值越小，两个文档越相似。in,

represents the weight that maps word i in document Q to word j in document _Ds , />

Indicates the distance between word i in document Q and word j in document D _s ; it is worth noting that the i-th word in document Q corresponds to the sum of the weight values of all words in a document of D _s equal to />

, in the same way, the sum of the weight values of all words in the document Q mapped to the jth word of the document in D _s is equal to />

, where the smaller the f-value, the more similar the two documents are.

S244. Calculate the similarity between the document Q and the document D _s :

，根据文档Q和文档D_s的相似度得到小于阈值/>

的k个RPA项目资产文档以及k个项目对应的知识图谱和流程图。S245, setting a similarity threshold

, according to the similarity between document Q and document D _s , it is less than the threshold />

The k RPA project asset documents and the knowledge maps and flowcharts corresponding to the k projects.

In a further embodiment, refer to Fig. 2 and Fig. 3, Fig. 2 is the user demand knowledge map of the Agricultural Bank of China's online banking running water downloading constructed by the embodiment of the present invention, and Fig. 3 is the RPA asset library project B company's online banking running water downloading knowledge constructed by the embodiment of the present invention Atlas. The user's demand documents include Agricultural Bank of China online banking records, etc., which can be matched to similarly completed project documents in the RPA asset library, such as company B's online banking records, including CCB online banking records and ICBC online banking records.

S3. Construct a query knowledge map according to the user requirement document, and set a start node. As shown in Figure 2, the user's demand is the download of the ABC online banking flow, and includes the sub-steps of the ABC online banking U-Shield login, the ABC online banking flow export, and the ABC online banking flow data conversion.

S4. Search and match the subgraphs to be matched in the query knowledge graph, obtain the final best matching graph, and return the corresponding flow chart.

S41. Use the TALE approximate large graph matching tool to search and query the best possible matching result of the knowledge graph in the subgraph to be matched.

S42. If the best matching result is found in step S41, return the matched candidate subgraph and the flow chart corresponding to the matched candidate subgraph; if no search is found, execute step S43.

S43. According to the attribute of the edge of the current start node, divide the graph into a subgraph set for the inclusion relationship, and repeat steps S41 and S42. At this time, the start node is updated as the tail entity of the current node until the best matching result or subgraph is obtained. The collection ends when it is empty.

The present invention also proposes a flow chart generation device based on semantic similarity and subgraph matching, including:

processor;

a memory on which is stored a computer program executable on said processor;

Wherein, when the computer program is executed by the processor, a flowchart generation method based on semantic similarity and subgraph matching is implemented.

In the technical solution proposed by the present invention, the WMD algorithm is used to measure the similarity between the user demand document and the RPA project asset library document, and find items similar to the current user demand to reduce the scope of further search using subgraph matching and improve efficiency.

Combining the method of graph partitioning and subgraph matching, iteratively searches for the best matching subgraph. Use the method of fuzzy subgraph matching to search the user demand knowledge map in the RPA project asset knowledge map for the best matching subgraph. Fuzzy subgraph matching allows some nodes to be mismatched and some edges to be missing. After recommending the relevant flow chart , the flow chart can be amended manually. When the search cannot be found, use the idea of edge division of graph division and consider the characteristics of the user demand knowledge graph, limit the relationship edge to the "containment" relationship for subgraph division, and then search for the best matching subgraph.

Using the method of semantic similarity and subgraph matching to jointly generate a flow chart: first, use semantic similarity to conduct the first round of search, get historical items that are relatively similar to user needs, and reduce the second round of subgraph matching search space; secondly, use fuzzy subgraph matching Graph matching conducts a second round of search, finds historical items that are similar in semantics and structure to user needs as the best matching item, and finds the corresponding flow chart.

The above description of the disclosed embodiments is provided to enable any person 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 present invention will not 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 (8)

1. A flow chart generation method based on semantic similarity and subgraph matching, characterized in that, comprising the following steps: S1. Obtain user requirement documents and RPA project asset library documents; S2. Calculate the semantic similarity between the user requirement document and the RPA project asset library document, and arrange top-k subgraphs to be matched according to the semantic similarity from high to low; S3. Construct a query knowledge map according to the user requirement document, and set a start node; S4. Search and match the subgraphs to be matched in the query knowledge graph, obtain the final best matching graph, and return the corresponding flow chart. 2. A method for generating a flow chart based on semantic similarity and subgraph matching according to claim 1, wherein step S2 is specifically: S21. Requirements document Q for users and user documents of RPA project asset library

Perform word segmentation and remove stop words; S22. Use the frequency of words in the text to describe the document, and combine the user's demand document Q and the user document of the RPA project asset library

Expressed as a one-dimensional vector; S23. According to the one-dimensional vector obtained in S22, learn each word of the user requirement document Q and the user document of the RPA project asset library

The embedding vector of each word in ; S24, use the WMD algorithm to calculate the user requirement document Q and the user document of the RPA project asset library

similarity between texts. 3. A flow chart generation method based on semantic similarity and subgraph matching according to claim 2, characterized in that in step S21, the word segmentation component jieba is used to analyze the user's demand document Q and the RPA project asset library user document

Participate. 4. A flow chart generation method based on semantic similarity and subgraph matching according to claim 2, characterized in that, in step S22, the method of normalized BOW bag-of-words model is used to represent Q and

. 5. a kind of flow chart generation method based on semantic similarity and subgraph matching according to claim 2, it is characterized in that, in step S23, use word2vec method to learn each word of user demand document Q and RPA project asset library user document

Embedding vectors for each word in . 6. A method for generating a flow chart based on semantic similarity and subgraph matching according to claim 2, wherein step S24 is specifically: S241, document Q and document

The number of occurrences of the words in the document Q is normalized to calculate the word frequency of the i-th vocabulary in the document Q/>

, document />

The word frequency of the jth vocabulary in is />

:

Among them, m and n are the number of words in document Q and document D _s respectively,

, />

Respectively represent the number of occurrences of the i-th word and j-th word in document Q and document D _s ; S242. Calculate the Euclidean distance between word i from document Q and word j from document D _s as C _i,j :

in,

and />

Embedding vectors learned for words i and j; S243. Using a dynamic programming algorithm to solve the WMD distance between the document Q and the document D _s :

where f is the WMD distance between document Q and document D _s ,

represents the weight that maps word i in document Q to word j in document _Ds , />

Indicates the distance between word i in document Q and word j in document D _s ; it is worth noting that the i-th word in document Q corresponds to the sum of the weight values of all words in a document of D _s equal to />

, in the same way, the sum of the weight values of all words in the document Q mapped to the jth word of the document in D _s is equal to />

; S244. Calculate the similarity between the document Q and the document D _s :

S245, setting a similarity threshold

, according to the similarity between document Q and document D _s , it is less than the threshold />

The k RPA project asset documents and the knowledge maps and flowcharts corresponding to the k projects. 7. A method for generating a flow chart based on semantic similarity and subgraph matching according to claim 1, wherein step S4 is specifically: S41. Use the TALE approximate large graph matching tool to search and query the best possible matching result of the knowledge graph in the subgraph to be matched; S42. If the best matching result is found in step S41, the matched candidate subgraph and the flow chart corresponding to the matched candidate subgraph are returned. If no search is found, step S43 is executed; S43. According to the attribute of the edge of the current start node, divide the graph into a subgraph set for the inclusion relationship, and repeat steps S41 and S42. At this time, the start node is updated as the tail entity of the current node until the best matching result or subgraph is obtained. The collection ends when it is empty. 8. A flow chart generation device based on semantic similarity and subgraph matching, characterized in that the device comprises: processor; a memory on which is stored a computer program executable on said processor; Wherein, when the computer program is executed by the processor, a flow chart generation method based on semantic similarity and subgraph matching according to any one of claims 1 to 7 is implemented.

Images