A Multi-Heterogeneous Data Fusion Method for Project Management Systems

Abstract

This application relates to the field of data fusion technology and proposes a method for fusing multi-source heterogeneous data in project management systems. The method includes: extracting all multi-source heterogeneous data from the project management system within a data analysis cycle, wherein the multi-source heterogeneous data consists of file data; constructing the topic-keyword consistency and topic word association volatility of the file data to determine the important topic word set of the file data; and combining the topic-keyword consistency and topic word association volatility to cluster all file data in the multi-source heterogeneous data to obtain the data fusion result. This application aims to solve the problem that single-scale text features cannot accurately capture local topic features of data, leading to inaccurate data fusion results.

Description

Technical Field

[0001] This application relates to the field of data fusion technology, specifically to a method for fusion of diverse heterogeneous data for project management systems. Background Technology

[0002] Project management systems enable sequential control of multiple processes within a project and parallel collaborative management of multiple projects, supporting project teams in efficiently achieving project goals and possessing multi-objective combination management capabilities. However, the data uploaded by various departments within the system exhibits diverse and heterogeneous characteristics, making it difficult to fully extract the effective information hidden within the data, and the correlation between data with the same attributes cannot be effectively identified and utilized. This, in turn, affects the integrity of project data and its practical application value, necessitating a solution through multi-source heterogeneous data fusion and aggregation methods.

[0003] Existing methods for fusing diverse heterogeneous data mainly include database reading and ETL tools. Database reading methods complete data migration through cross-database read / write operations. However, when dealing with heterogeneous data of diverse structures and sources, this often results in fragmented storage of data with similar attributes, hindering effective merging. ETL tools can establish mapping relationships between data features and target structures to achieve data fusion, but when processing large-scale heterogeneous data, the semi-automatic processing mode is time-consuming and cannot meet real-time fusion requirements. Therefore, existing methods often extract semantic features at a single text granularity to construct data associations, making it difficult to capture local thematic features of the data at multiple scales. They tend to overlook key keywords with high local saliency but low global weight, leading to insufficient data representation capabilities and reduced data fusion depth and value mining effectiveness. Summary of the Invention

[0004] This application provides a multi-source heterogeneous data fusion method for project management systems to address the problem that single-scale text features cannot accurately capture local thematic features of data, leading to inaccurate data fusion results. The specific technical solution adopted is as follows:

[0005] One embodiment of this application provides a method for fusing diverse heterogeneous data in a project management system, the method comprising the following steps:

[0006] Extract all multi-dimensional heterogeneous data from the project management system within a data analysis cycle and standardize the format. The multi-dimensional heterogeneous data consists of file data.

[0007] Each file in the multivariate heterogeneous data is segmented and sentence-by-sentence. The topic set and keyword set of the file data, sentences in the file data, and paragraphs in the file data are extracted. Based on the degree of overlap between the topic set and the keyword set, the topic-keyword consistency of the file data is constructed. The topic-keyword consistency is used to evaluate the consistency difference of the topic of the file data at different scales. The high-frequency encoding vectors of each topic word in the multivariate heterogeneous data are extracted. Based on the similarity of the high-frequency encoding vectors of different topic words, the association probability of topic words and the topic word association volatility of the file data are calculated. The topic word association volatility is used to evaluate the uniformity of semantic association distribution among topic words in the file data. Combining the topic-keyword consistency and the frequency of the topic words in the file data, sentences in the file data, and paragraphs in the file data, the importance of topic words is constructed, and then the important topic word set of the file data is determined.

[0008] Based on the degree of overlap in the important keyword sets of two different document data, the difference in topic-keyword consistency, and the difference in keyword association volatility, clustering is performed on all document data in the multivariate heterogeneous data to obtain the data fusion results of the multivariate heterogeneous data.

[0009] Furthermore, the method for constructing the topic-keyword consistency of the file data is as follows:

[0010] Based on the degree of overlap between the topic set and keyword set of the document data, the degree of overlap between the keyword set and topic set of sentences in the document data, and the degree of overlap between the topic set and keyword set of paragraphs in the document data, the first similarity, second similarity, and third similarity of the document data are determined respectively.

[0011] Based on the first, second, and third similarity scores of the file data, a topic-keyword consistency is constructed. The topic-keyword consistency is negatively correlated with the second and third similarity scores and positively correlated with the first similarity score.

[0012] Furthermore, the first, second, and third similarities of the file data are all determined using Jaccard similarity.

[0013] Furthermore, the specific calculation of the association probability of the topic terms is as follows:

[0014] The cosine similarity of the high-frequency encoding vectors of two different topic words is used as the semantic similarity of the two different topic words. Based on the semantic similarity, the semantically similar topic words of each topic word are determined. The ratio of the frequency of a topic word being a semantically similar topic word to other topic words to the total number of topic words is recorded as the association probability of the topic word.

[0015] Furthermore, the specific method for determining the semantically similar keywords is as follows:

[0016] The mean of the semantic similarity of all different topic terms in the file data is recorded as the semantic similarity threshold of the file data;

[0017] For the same file data, two different topic terms with a semantic similarity greater than the semantic similarity threshold are both recorded as semantically similar topic terms of each other.

[0018] Furthermore, the volatility of the keyword association in the file data is: the normalized value of the information entropy of the association probability of all keywords in the file data.

[0019] Furthermore, the method for constructing the importance level of the keywords is as follows:

[0020] The product of the mean result of the consistency between the subject and keyword in the file data containing the subject and the volatility of the subject association is multiplied by the probability of association of the subject and is recorded as the contribution weight of the subject.

[0021] The number of times a keyword appears in the subject set of document data, sentences in document data, and paragraphs in document data is denoted as the multi-scale frequency of the keyword.

[0022] The importance of the keywords is the product of the frequency of the keywords in the document data and the multi-scale frequency and contribution weight of the keywords.

[0023] Furthermore, the method for determining the set of important keywords in the document data is as follows:

[0024] The average importance of all keywords in the document data is recorded as the importance threshold. The set of all keywords in the document data whose importance is greater than the importance threshold is recorded as the important keyword set of the document data.

[0025] Furthermore, the method for clustering all file data in the multivariate heterogeneous data based on the degree of overlap in the important keyword sets of two different file data, the difference in topic-keyword consistency, and the difference in keyword association volatility, to obtain the data fusion result of the multivariate heterogeneous data, includes the following specific methods:

[0026] The feature distance between two different document datasets is calculated based on the degree of overlap in the important keyword sets, the difference in topic-keyword consistency, and the difference in keyword association volatility.

[0027] The feature distance between two different file data is used as the metric distance of the clustering algorithm to cluster all file data in the multivariate heterogeneous data, and file data clusters are obtained. File data within the same file data cluster are successfully matched file data in the multivariate heterogeneous data.

[0028] Furthermore, the method for determining the feature distance between the two different file data is as follows:

[0029] The difference in Jaccard similarity between the number 1 and the important topic word sets of two different document data is denoted as the first distance between the two different document data. The absolute value of the difference in topic-keyword consistency between the two different document data is denoted as the second distance between the two different document data. The absolute value of the difference in topic word association volatility between the two different document data is denoted as the third distance between the two different document data.

[0030] The positive correlation results of the first, second, and third distances between two different file data are denoted as the feature distance between the two different file data.

[0031] The beneficial effects of this application are:

[0032] This application considers that at the document level, paragraph level, and sentence level, the higher the overlap between the topic set and keyword set at the same scale, the more significant the consistency between the overall topic and core keywords at the corresponding scale. Based on the overlap between the topic set and keyword set of the document data, sentences within the document data, and paragraphs within the document data, this application evaluates the consistency differences of the document data's topic at different scales, obtains the topic-keyword consistency of the document data, and evaluates the uniformity of semantic association distribution among topic words within the document data, obtaining the volatility of topic word associations in the document data. The greater the volatility of topic word associations in the document data, the more significant the consistency between the topic sets and keyword sets at each scale. The more uneven the distribution of semantic associations among keywords, the more important the keywords in the document data are evaluated by combining three scales: document level, paragraph level, and sentence level. Based on the keywords with higher importance in the document data, the important keyword set of the document data is determined. Finally, based on the degree of overlap of important keyword sets, the difference in topic-keyword consistency, and the difference in keyword association volatility between two different document data sets, all document data in the multi-dimensional heterogeneous data are clustered to obtain the data fusion results of multi-dimensional heterogeneous data. This solves the problem that single-scale text features cannot accurately capture local topic features of the data, leading to inaccurate data fusion results. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1This is a schematic flowchart of a multi-heterogeneous data fusion method for a project management system provided in one embodiment of this application. Detailed Implementation

[0035] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0036] Please see Figure 1 The diagram illustrates a flowchart of a multi-source heterogeneous data fusion method for a project management system, provided in one embodiment of this application. The method includes the following steps:

[0037] Step S001: Extract all multi-dimensional heterogeneous data from the project management system within a data analysis cycle and unify the format. The multi-dimensional heterogeneous data consists of file data.

[0038] The data analysis period is preset, and in this embodiment, the data analysis period is set to one calendar day.

[0039] This involves extracting all diverse and heterogeneous data received by the project management system within a data analysis cycle, unifying the format of this data, and converting data from different sources and with different structures into text data with unified attributes. The diverse and heterogeneous data consists of file data. Specifically, the format unification process in this embodiment is as follows:

[0040] 1. Set data source type mapping rules: map relational data such as CSV and XLSX to table structure descriptions; map JSON and XML data to key-value pair descriptions; map unstructured data such as images, audio, and video to filename-path-basic metadata descriptions.

[0041] 2. Set up a data flattening template: For unstructured data such as images, audio, and video, extract the content text in advance using optical character recognition or speech recognition technology, and then concatenate the extracted content text with the file name, path, and basic metadata to form the flattening description text. Structured data is flattened in the format of "field[field name]: value[field value]"; unstructured data is flattened in the format of "[data source type], data ID: [identifier], content summary: [flattening description text]"; when flattening JSON and XML data, remove common key names such as id and key, and only retain key-value pairs with business meaning.

[0042] The data types of the diverse and heterogeneous data are obtained from the API interface for data transmission to the project management system. The data types of the diverse and heterogeneous data include relational data such as CSV and XLSX, as well as JSON and XML data.

[0043] In summary, by following the mapping rules and flattening templates described above, structured data is flattened into natural language text, and unstructured data is converted into metadata and text descriptions, forming a plain text dataset in a unified format.

[0044] Furthermore, data cleaning is performed on the diverse and heterogeneous data to eliminate errors and inconsistencies. Specifically, the data cleaning process in this embodiment is as follows: length filtering is used to remove excessively short or long text from the diverse and heterogeneous data with uniform format, and garbled characters are eliminated through garbled character detection.

[0045] Among them, length filtering and garbled character detection are well-known technologies in this field, and will not be elaborated here.

[0046] At this point, we have obtained diverse and heterogeneous data with a unified format.

[0047] Step S002: Segment and sentence-by-sentence the file data in the multivariate heterogeneous data, extract the topic set and keyword set of the file data, sentences in the file data, and paragraphs in the file data. Based on the degree of overlap between the topic set and the keyword set, construct the topic-keyword consistency of the file data. The topic-keyword consistency is used to evaluate the consistency difference of the topic of the file data at different scales. Extract the high-frequency encoding vector of each topic word in the multivariate heterogeneous data. Based on the similarity of the high-frequency encoding vectors of different topic words, calculate the association probability of topic words and the topic word association volatility of the file data. The topic word association volatility is used to evaluate the uniformity of semantic association distribution among topic words in the file data. Combine the topic-keyword consistency and the frequency of the topic words in the file data, sentences in the file data, and paragraphs in the file data to construct the importance of topic words, and then determine the important topic word set of the file data.

[0048] Data fusion of diverse heterogeneous data can be equivalent to a data matching process. The core is to categorize paired heterogeneous data into different classes and then merge paired heterogeneous data within the same class. Project management systems focus on planning, executing, monitoring, and completing projects. The degree of correlation between data in different processes varies significantly. For example, the project planning phase only clarifies the execution object but lacks specific execution details, while the data in the monitoring and execution phases are aligned across most detailed dimensions. Based on these characteristics, if the same file data is segmented into text data of different scales, the similarity of keywords under the multi-scale data will also change accordingly.

[0049] For any file in a multi-dimensional heterogeneous dataset: set the paragraph delimiter to a newline character and a carriage return character, and the minimum paragraph length is... The maximum sentence length is [number] characters; the sentence segmentation identifier is set to a period, question mark, or other ending punctuation mark. Characters; set the minimum word length for filtering words to be [characters]. Each character. Based on a project management domain dictionary, the HanLP algorithm is used to process the text content of the file data by removing stop words, segmenting, and dividing it into sentences.

[0050] in, , , These represent the first preset quantity, the second preset quantity, and the third preset quantity, respectively. The first preset quantity should be an integer greater than or equal to 10 and less than or equal to 50, the second preset quantity should be an integer greater than or equal to 100 and less than or equal to 500, and the third preset quantity should be 1 or 2. In this embodiment, the values ​​for the first preset quantity, the second preset quantity, and the third preset quantity are 20, 200, and 1, respectively. The project management domain dictionary is a collection of professional terms specifically built for project management scenarios. It is essentially a structured vocabulary list that includes frequently occurring professional concepts, terms, abbreviations, and industry terms throughout the entire project management lifecycle, and is different from general Chinese dictionaries or ordinary stop word lists.

[0051] The LDA algorithm is used to obtain the topic set of each file in the multivariate heterogeneous data, thus extracting the topic of the file data. The TextRank algorithm is used to extract the keywords and keyword weights of each file in the multivariate heterogeneous data, constructing a keyword set for the file data. HanLP text processing is used to segment each file in the multivariate heterogeneous data into sentences, obtaining sentences from the file data. The LDA and TextRank algorithms are used to extract the topic set and keyword set of the sentences from the file data, respectively. Based on the paragraph segmentation identifier, the segments of each file in the multivariate heterogeneous data are obtained, and the paragraphs from the file data are obtained. The LDA and TextRank algorithms are used to extract the topic set and keyword set of the paragraphs from the file data, respectively. A distributed word vector model is employed to obtain the high-frequency encoding vector of each topic word in the multivariate heterogeneous data.

[0052] In this embodiment, the number of topics for the LDA algorithm is set to 20, the number of iteration rounds is set to 50, the 20 keywords with the highest weights are selected to construct a keyword set, the window size of the one-hot encoding technique is set to 5, and the damping coefficient is set to 0.85. The LDA algorithm, TextRank algorithm and one-hot encoding technique are all well-known technologies and will not be described in detail. The distributed word vector model can be Word2Vec or BERT.

[0053] For any file in the multivariate heterogeneous data, the Jaccard similarity between the topic set and the keyword set of the file is denoted as the first similarity of the file; the Jaccard similarity between the keyword set and the topic set of sentences in the file is denoted as the second similarity of the file; and the Jaccard similarity between the topic set and the keyword set of paragraphs in the file is denoted as the third similarity of the file. Based on the first, second, and third similarities of the file, the topic-keyword consistency of the file is constructed. The topic-keyword consistency is negatively correlated with the second and third similarities and positively correlated with the first similarity.

[0054] It is understood that the positive and negative correlations in this application refer to the relationship between the independent and dependent variables. A positive correlation means that the dependent variable increases (decreases) as the independent variable increases (decreases), and can be an additive or multiplicative relationship. A negative correlation means that the dependent variable decreases (increases) as the independent variable increases (decreases), and can be an inverse relationship or a subtractive relationship.

[0055] Preferably, as an embodiment of this application, the ratio of the sum of the second similarity and the third similarity of the file data to the first similarity is denoted as the file data similarity ratio, and the difference between the number 1 and the file data similarity ratio is denoted as the subject-keyword consistency of the file data.

[0056] It is important to understand that the subject set and keyword set of file data will always contain the same words, so the first similarity of the file data cannot be 0.

[0057] At the document, paragraph, and sentence levels, the higher the overlap between the topic set and keyword set at the same level, the more significant the consistency between the overall topic and core keywords at that level. Differences will exist between the topic set and keyword set at different levels, ensuring that the similarity ratio of the document data is greater than 0 and less than 1. The topic-keyword consistency of document data is used to evaluate the consistency differences of the topic at different levels, effectively distinguishing the characteristics of documents at different stages of project management. For example, the planning stage has a clear macro-theme but insufficient micro-details, while the execution and monitoring stages have a high degree of consistency between macro and micro aspects.

[0058] When the keywords in the document data are more important, they appear more frequently in the keyword set at different scales. Simultaneously, the high-frequency encoding vectors of different keywords corresponding to different expressions of the same meaning show a high degree of correlation. Examples of keywords with different expressions of the same meaning include "vehicle" and "car."

[0059] The cosine similarity of the high-frequency encoding vectors of two different topic words is used as the semantic similarity of the two different topic words. The mean of the semantic similarity of all different topic words in the file data is recorded as the semantic similarity threshold of the file data. For the same file data, two different topic words with semantic similarity greater than the semantic similarity threshold are both recorded as semantically similar topic words of each other. The ratio of the frequency of a topic word being a semantically similar topic word to the total number of topic words is recorded as the association probability of the topic word. The normalized value of the information entropy of the association probability of all topic words in the file data is recorded as the topic word association volatility of the file data.

[0060] The calculation of cosine similarity and information entropy are well-known techniques and will not be elaborated further.

[0061] It is understandable that semantically similar keywords are different keywords that correspond to different expressions of the same meaning.

[0062] In one embodiment of this application, the value of the logarithm with the number 2 as the base and the total number of keywords as the argument is taken as the denominator, and the information entropy of the association probability of all keywords in the file data is taken as the numerator as the fraction value, which is denoted as the keyword association volatility of the file data.

[0063] When performing fractional or logarithmic calculations, to prevent the calculation from being meaningless due to a denominator of 0 or a logarithmic argument of 0, a very small positive number is added to the denominator or logarithmic argument as a smoothing parameter adjustment factor. In this embodiment, the value of the very small positive number is 0.00001.

[0064] The greater the volatility of the subject associations in the document data, the more uneven the distribution of semantic associations among the subject terms within the document data.

[0065] Under standard project management processes, as planning progresses towards execution, the consistency of the subject-keyword profile and the volatility of the subject-term association in the corresponding document data should gradually decrease.

[0066] The product of the mean result of the topic-keyword consistency and topic association volatility of the document data containing the topic term and the association probability of the topic term is recorded as the contribution weight of the topic term; the number of times the topic term appears in the document data, sentences in the document data, and paragraphs in the document data is recorded as the multi-scale frequency of the topic term; and the product of the frequency of the topic term in the document data and the multi-scale frequency and contribution weight of the topic term is recorded as the importance of the topic term.

[0067] It is understood that positive correlation processing is applied to the frequency of keyword occurrences in the document data, the multi-scale frequency of keyword occurrences, and the contribution weight, ensuring that these factors are positively correlated with the importance of the keyword. It is also understood that the positive correlation in this application refers to the relationship between the independent and dependent variables. The independent variables are the frequency of keyword occurrences in the document data, the multi-scale frequency of keyword occurrences, and the contribution weight; the dependent variable is the importance of the keyword. A positive correlation means that the dependent variable increases (decreases) as the independent variable increases (decreases), and this can be an additive or multiplicative relationship.

[0068] Preferably, as an embodiment of this application, the positive correlation between the frequency of a keyword in the document data and the multi-scale frequency and contribution weight of the keyword is recorded as the importance of the keyword.

[0069] The greater the contribution weight of a keyword, the greater the probability of association and semantic similarity between the keyword and other keywords in the document data. The multi-scale frequency of keywords is used to evaluate the comprehensiveness of the semantic feature coverage of keywords. The calculation of the importance of keywords combines three scales: document level, paragraph level, and sentence level, to evaluate the importance of keywords in the document data.

[0070] The average importance of all keywords in the document data is recorded as the importance threshold. The set of all keywords in the document data whose importance is greater than the importance threshold is recorded as the important keyword set of the document data.

[0071] It should be noted that if all keywords in the file are of equal importance, resulting in an empty set of keywords with values ​​greater than the average, then the top 5 keywords based on importance will be selected to form the important keyword set.

[0072] This completes the acquisition of a set of key keywords for the file data.

[0073] Step S003: Based on the degree of overlap of important keyword sets, the difference in topic-keyword consistency, and the difference in keyword association volatility between two different file data, cluster all file data in the multivariate heterogeneous data to obtain the data fusion results of the multivariate heterogeneous data.

[0074] The difference in Jaccard similarity between the number 1 and the important topic word sets of two different document data is denoted as the first distance between the two different document data. The absolute value of the difference in topic-keyword consistency between the two different document data is denoted as the second distance between the two different document data. The absolute value of the difference in topic word association volatility between the two different document data is denoted as the third distance between the two different document data. The positive correlation result of the first, second, and third distances between the two different document data is denoted as the feature distance between the two different document data.

[0075] Preferably, as an embodiment of this application, the product of the mean of the second and third distances of two different file data and the first distance is recorded as the feature distance between the two different file data.

[0076] The feature distance between two different file data is used as the metric distance for the clustering algorithm to cluster all file data in the multivariate heterogeneous data and obtain file data clusters.

[0077] It is understandable that file data within the same file data cluster is the file data that has been successfully matched in the multi-dimensional heterogeneous data.

[0078] This completes the data fusion of diverse and heterogeneous data.

[0079] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A method for fusing diverse heterogeneous data in a project management system, characterized in that, The method includes the following steps: Extract all multi-dimensional heterogeneous data from the project management system within a data analysis cycle and standardize the format. The multi-dimensional heterogeneous data consists of file data. Each file in the multivariate heterogeneous data is segmented and sentence-by-sentence. The topic set and keyword set of the file data, sentences in the file data, and paragraphs in the file data are extracted. Based on the degree of overlap between the topic set and the keyword set, the topic-keyword consistency of the file data is constructed. The topic-keyword consistency is used to evaluate the consistency difference of the topic of the file data at different scales. The high-frequency encoding vectors of each topic word in the multivariate heterogeneous data are extracted. Based on the similarity of the high-frequency encoding vectors of different topic words, the association probability of topic words and the topic word association volatility of the file data are calculated. The topic word association volatility is used to evaluate the uniformity of semantic association distribution among topic words in the file data. Combining the topic-keyword consistency and the frequency of the topic words in the file data, sentences in the file data, and paragraphs in the file data, the importance of topic words is constructed, and then the important topic word set of the file data is determined. Based on the degree of overlap in the important keyword sets of two different file data, the difference in topic-keyword consistency, and the difference in keyword association volatility, all file data in the multivariate heterogeneous data are clustered to obtain the data fusion results of the multivariate heterogeneous data; The method for establishing topic-keyword consistency in the file data is as follows: Based on the degree of overlap between the topic set and keyword set of the document data, the degree of overlap between the keyword set and topic set of sentences in the document data, and the degree of overlap between the topic set and keyword set of paragraphs in the document data, the first similarity, second similarity, and third similarity of the document data are determined respectively. Based on the first similarity, second similarity, and third similarity of the file data, a topic-keyword consistency of the file data is constructed. The topic-keyword consistency is negatively correlated with the second similarity and third similarity, and positively correlated with the first similarity. The specific calculation of the association probability of the topic terms is as follows: The cosine similarity of the high-frequency encoding vectors of two different topic words is used as the semantic similarity of the two different topic words. Based on the semantic similarity, the semantically similar topic words of each topic word are determined. The ratio of the frequency of a topic word being a semantically similar topic word to other topic words to the total number of topic words is recorded as the association probability of the topic word. The volatility of the keyword association in the file data is the normalized value of the information entropy of the association probability of all keywords in the file data.

2. The method for multi-source heterogeneous data fusion in a project management system according to claim 1, characterized in that, The first, second, and third similarities of the file data are all determined using Jaccard similarity.

3. The method for multi-source heterogeneous data fusion in a project management system according to claim 1, characterized in that, The specific method for determining the semantically similar keywords is as follows: The mean of the semantic similarity of all different topic terms in the file data is recorded as the semantic similarity threshold of the file data; For the same file data, two different topic terms with a semantic similarity greater than the semantic similarity threshold are both recorded as semantically similar topic terms of each other.

4. The method for multi-source heterogeneous data fusion in a project management system according to claim 1, characterized in that, The method for constructing the importance of the keywords is as follows: The product of the mean result of the consistency between the subject and keyword in the file data containing the subject and the volatility of the subject association is multiplied by the probability of association of the subject and is recorded as the contribution weight of the subject. The number of times a keyword appears in the subject set of document data, sentences in document data, and paragraphs in document data is denoted as the multi-scale frequency of the keyword. The importance of the keywords is the product of the frequency of the keywords in the document data and the multi-scale frequency and contribution weight of the keywords.

5. The method for multi-source heterogeneous data fusion in a project management system according to claim 1, characterized in that, The method for determining the set of important keywords in the document data is as follows: The average importance of all keywords in the document data is recorded as the importance threshold. The set of all keywords in the document data whose importance is greater than the importance threshold is recorded as the important keyword set of the document data.

6. The method for multi-source heterogeneous data fusion in a project management system according to claim 1, characterized in that, The method for clustering all file data in multivariate heterogeneous data based on the degree of overlap in important keyword sets, differences in topic-keyword consistency, and differences in keyword association volatility between two different file data sets, and obtaining the data fusion results of multivariate heterogeneous data, includes the following specific methods: The feature distance between two different document datasets is calculated based on the degree of overlap in the important keyword sets, the difference in topic-keyword consistency, and the difference in keyword association volatility. The feature distance between two different file data is used as the metric distance of the clustering algorithm to cluster all file data in the multivariate heterogeneous data, and file data clusters are obtained. File data within the same file data cluster are successfully matched file data in the multivariate heterogeneous data.

7. A method for multi-source heterogeneous data fusion in a project management system according to claim 6, characterized in that, The method for determining the feature distance between the two different file data is as follows: The difference in Jaccard similarity between the number 1 and the important topic word sets of two different document data is denoted as the first distance between the two different document data. The absolute value of the difference in topic-keyword consistency between the two different document data is denoted as the second distance between the two different document data. The absolute value of the difference in topic word association volatility between the two different document data is denoted as the third distance between the two different document data. The positive correlation results of the first, second, and third distances between two different file data are denoted as the feature distance between the two different file data.

Images