Heterogeneous information fusion method and system based on graph neural network

Abstract

The invention discloses a heterogeneous information fusion method and system based on a graph neural network, and relates to the technical field of heterogeneous information fusion. The method comprises the following steps: collecting information data, preprocessing the information data, and taking the preprocessed data as basic data; setting conversion standards of different types of basic data, and converting the basic data according to the conversion standards to obtain elementary data; corresponding entities are extracted and recognized according to the elementary data and serve as standard entity objects; acquiring acquisition information of elementary data, judging data quality of the elementary data according to the acquisition information and the standard entity object, and matching according to the data quality to obtain medium data; and forming a heterogeneous information graph according to the medium data, and selecting a graph neural network model to process the heterogeneous information graph to obtain a heterogeneous information fusion result. According to the invention, the accuracy of heterogeneous information fusion based on the graph neural network is improved.

Description

Technical Field

[0001] The present application relates to the technical field of heterogeneous information fusion, and in particular to a heterogeneous information fusion method and system based on graph neural network. Background Art

[0002] Heterogeneous information fusion refers to the process of effectively integrating information data from different data sources, formats and types. Heterogeneous information fusion makes information data more comprehensive and accurate. This comprehensiveness and accuracy help to explore deeper connections and laws between information data, thereby leveraging greater data value. Moreover, heterogeneous information fusion can break down data barriers and promote data interconnection between different organizations, departments or systems. This helps to achieve data sharing and exchange and improve the utilization rate of data resources. In the process of heterogeneous information fusion, there are differences in the quality of information from different sources, and the quality of information data is different, so the quality of the fused information data will also be affected. This impact may cause the fused information data to fail to accurately reflect the actual situation, thereby reducing the effect of information data fusion. At present, data quality differences are reduced by data cleaning, such as cleaning missing values, duplicate data, and erroneous data. But obviously, in addition to these basic problems, the quality of information data is considered to have certain differences because of data accuracy, which seriously reduces the effect of heterogeneous information fusion and reduces the accuracy of the fused information data results. Summary of the invention

[0003] The purpose of the present invention is to provide a heterogeneous information fusion method and system based on graph neural network to solve the problems raised in the above background technology.

[0004] In the first aspect, the present application provides a heterogeneous information fusion method based on graph neural network, which adopts the following technical solution:

[0005] Collect information data, pre-process the information data, and use the pre-processed data as basic data;

[0006] Set conversion standards for different categories of basic data, convert the basic data according to the conversion standards, and obtain primary data;

[0007] According to the elementary data extraction, the corresponding entities are identified and used as standard entity objects;

[0008] Obtaining the collection information of primary data, judging the data quality of the primary data according to the collection information and standard entity objects, and obtaining the intermediate data according to the data quality matching;

[0009] A heterogeneous information graph is formed according to the medium data, and a graph neural network model is selected to process the heterogeneous information graph to obtain the heterogeneous information fusion result.

[0010] Preferably, the steps of obtaining the collection information of the primary data, judging the data quality of the primary data according to the collection information and the standard entity object, and obtaining the intermediate data according to the data quality matching are specifically as follows:

[0011] Acquire collection information of primary data, wherein the collection information includes data generation time and collection related information;

[0012] The collection-related information includes the collection source, collection occasion and collection environment, and the accuracy of the primary data is analyzed based on the collection-related information and the standard entity object;

[0013] Obtain the effective duration of the primary data, and determine the validity of the primary data based on the data generation time and effective duration;

[0014] Obtain the data type of the primary data, and determine the completeness of the primary data based on the data generation time and collection environment;

[0015] The data weight ratios of accuracy, validity and completeness are set respectively, and the data quality is calculated according to the accuracy, validity, completeness and the corresponding data weight ratios;

[0016] Set data quality standards, match data that meet the data quality standards from primary data and record them as medium data.

[0017] Preferably, the step of analyzing the accuracy of the primary data based on the collected association information and the standard entity object is specifically:

[0018] According to the source of the collection, the original outputter of the primary data is determined to determine whether the original outputter has subjective consciousness;

[0019] If the original outputter has no subjective consciousness, then the historical output records of the original outputter are obtained, the data accuracy in the historical output records is extracted, and the data accuracy is used as the accuracy of the primary data;

[0020] If the original outputter has subjective consciousness, then the data output habits of the original outputter are obtained, and the data subjectivity of the primary data is determined based on the data output habits and the collection occasions;

[0021] The emotional state of the original outputter is obtained by analyzing the data output habits, and the data emotional degree of the primary data is obtained according to the emotional state;

[0022] Obtain the information fusion requirements of standard entity objects, and combine the data subjectivity and data sentiment to obtain the accuracy of primary data.

[0023] Preferably, if the original outputter has subjective consciousness, the step of obtaining the data output habit of the original outputter and determining the data subjectivity of the primary data according to the data output habit and the collection occasion is specifically as follows:

[0024] Collect words that describe subjective feelings and record them as subjective words;

[0025] Determine whether the primary data is describing objective facts. If the primary data is describing objective facts, then count the actual frequency AP of subjective words used in the primary data.

[0026] Extract the average frequency AB of subjective words used by the original outputter in the output habits of data;

[0027] Extract the occurrence time AF of the objective fact described by the primary data, obtain the contact time AJ of the original outputter's last contact with the objective fact-related information, and the contact frequency AL of the original outputter with the objective fact-related information;

[0028] Collect historical data collected at the collection site, and count the data accuracy rate AZ in the historical data collected;

[0029] The subjective degree correlation function is used to calculate the data subjective degree AG, subjective degree correlation function:

[0030]

[0031] Among them, a 1 、a 2 、a 3 is the scale factor and is greater than 0, and AS is the data generation time;

[0032] If the elementary vocabulary is not describing objective facts, the subjectivity of the data is at its maximum.

[0033] Preferably, the step of obtaining the emotional state of the original outputter according to the data output habit analysis and obtaining the data emotional degree of the primary data according to the emotional state is specifically as follows:

[0034] The emotional fluctuation degree of the original outputter is obtained, and the average emotional fluctuation degree is calculated according to the emotional fluctuation degree and recorded as the original emotional degree BY;

[0035] The words, symbols and expressions that express emotions are recorded as emotional words, and the actual usage frequency of emotional words in the elementary data is counted;

[0036] Extract the average frequency of emotional words used by the original outputter in the data output habits, calculate the difference between the actual frequency of emotional words used and the average frequency of emotional words used and record it as the actual emotional degree BS;

[0037] Extract objective facts of the original output in the primary data and find the fact type of the objective facts;

[0038] Find the output emotion fluctuation degree of the information outputter of the same type of information of the fact type, calculate the average value of the output emotion fluctuation degree and record it as the information emotion degree BX;

[0039] The emotional degree correlation function is used to calculate the data emotional degree BQ, emotional degree correlation function:

[0040]

[0041] Among them, b 1 、b 2 、b 3 is the scaling factor and is greater than 0.

[0042] Preferably, the step of obtaining the information fusion requirements of the standard entity object and obtaining the accuracy of the primary data by combining the data subjectivity and the data emotion is specifically as follows:

[0043] Obtain the information fusion requirements of standard entity objects and determine whether the information fusion requirements only require objective facts;

[0044] If the information fusion requirement is only objective facts, then a data subjectivity threshold is set, and the primary data that does not reach the data subjectivity threshold is screened and recorded as the first data. The data sentiment of the first data is used as the accuracy of the primary data, and the accuracy of the primary data other than the first data is 0;

[0045] If the information fusion requirement is not just objective facts, the first weight ratios of the data subjectivity and the data sentiment are set respectively, and the accuracy of the elementary data is calculated based on the data subjectivity, the data sentiment and the corresponding first weight ratios.

[0046] Preferably, the step of obtaining the effective duration of the primary data and determining the effectiveness of the primary data according to the data generation time and the effective duration is specifically as follows:

[0047] Get the effective duration of the elementary data, get the real time, and calculate the time difference between the data generation time and the real time;

[0048] Determine whether the time difference exceeds the valid duration. If the time difference exceeds the valid duration, the validity of the primary data is 0;

[0049] If the time difference does not exceed the effective duration, the elementary data of the same fact type are formed into a type data set;

[0050] Extract the actual meaning of the data in the type data set, and extract the actual meaning with the largest number as the standard meaning;

[0051] The actual meaning of the primary data is extracted, and the similarity between the actual meaning and the standard meaning is compared and recorded as the validity of the primary data.

[0052] Preferably, the step of obtaining the data type of the primary data and determining the completeness of the primary data in combination with the data generation time and the collection environment is specifically as follows:

[0053] Obtain the necessary attributes of the primary data, and determine whether the primary data is complete based on the necessary attributes. If the primary data is incomplete, the completeness of the primary data is 0;

[0054] If the primary data is complete, the data type of the primary data is obtained, the basic completeness standard of the data type is extracted, and the proportion of the primary data that meets the basic completeness standard is obtained and recorded as the completeness proportion;

[0055] Determine the degree of interference caused by the collection environment to the primary data based on the collection environment;

[0056] The completeness ratio of the completeness proportion and the completeness weight ratio of the interference degree are set respectively, and the completeness of the primary data is calculated according to the completeness ratio, the interference degree and the corresponding completeness weight ratio.

[0057] Preferably, the step of judging the interference degree of the collection environment to the primary data according to the collection environment is specifically:

[0058] Obtain the interference factors that affect the primary data, count the interference factors contained in the collection environment and record them as interference factors;

[0059] Obtain the interference range of the interference factor, obtain the actual interference value of the interference factor, and obtain the degree of the interference factor according to the actual interference value and the interference range;

[0060] The basic interference degree caused by the collection environment is obtained by summing up all interference factors and the corresponding interference factor degrees;

[0061] Count the transmission time and number of transmissions of the elementary data, and obtain the degree of transmission interference according to the transmission time and number of transmissions;

[0062] The basic interference degree and the transmission interference degree are superimposed and calculated to obtain the actual interference degree of the primary data.

[0063] In the second aspect, the present application provides a heterogeneous information fusion system based on graph neural network, which adopts the following technical solution:

[0064] A heterogeneous information fusion system based on graph neural network, comprising:

[0065] The data processing module collects information data, pre-processes the information data, and uses the pre-processed data as basic data;

[0066] The conversion standard module sets the conversion standards for different categories of basic data, and converts the basic data according to the conversion standards to obtain primary data;

[0067] The entity recognition module extracts and recognizes the corresponding entity based on the primary data and uses it as a standard entity object;

[0068] The data quality module obtains the collection information of the primary data, determines the data quality of the primary data based on the collection information and the standard entity object, and obtains the intermediate data based on the data quality matching;

[0069] The fusion result module forms a heterogeneous information graph based on the medium data, selects the graph neural network model to process the heterogeneous information graph, and obtains the heterogeneous information fusion result.

[0070] In summary, the present application includes at least one of the following beneficial technical effects:

[0071] 1. Analyze the accuracy of primary data based on the collection-related information and standard entity objects. The collection-related information includes the collection source, collection occasion, and collection environment. The validity of the primary data is determined based on the data generation time and the effective duration. The completeness of the primary data is determined based on the data generation time and the collection environment. The data quality is obtained comprehensively, and the data is matched according to the data quality standard. The accuracy, validity, and completeness of the data are obtained respectively based on the collection-related information of the data, and the data quality is comprehensively evaluated, which improves the comprehensiveness of heterogeneous information fusion based on graph neural networks.

[0072] 2. Determine whether the original outputter has subjective consciousness, and obtain different data accuracy according to whether there is subjective consciousness. If the original outputter has subjective consciousness, obtain the data output habits of the original outputter, and determine the data subjectivity of the primary data according to the data output habits and the collection occasions, and analyze the data emotion of the primary data according to the data output habits, so as to obtain the accuracy of the primary data in combination with the information fusion requirements of the standard entity object. Evaluate the accuracy of the data according to the subjectivity and emotion in the data, fully consider the impact of the original outputter on the data accuracy, and improve the accuracy of heterogeneous information fusion based on graph neural networks.

[0073] 3. The original emotional degree is calculated based on the degree of emotional fluctuation of the original outputter, and the actual emotional degree is obtained based on the use of emotional words, where emotional words are words, symbols and expressions that express emotions. At the same time, the output emotional situation of the information outputter of the same type of information as the objective facts of the original outputter is found to obtain the information emotional degree. Combined with the time of occurrence of the objective facts, the information emotional degree is evaluated from multiple aspects, not only considering the emotional impact of the original outputter, but also considering the emotional impact of the objective facts in the information data on the original outputter. Finally, the data emotional degree is confirmed, which improves the practicality of heterogeneous information fusion based on graph neural networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is a schematic diagram of the specific steps of an embodiment of a heterogeneous information fusion method based on a graph neural network of the present invention.

[0075] Figure 2 It is a module connection diagram of an embodiment of a heterogeneous information fusion system based on graph neural network of the present invention. DETAILED DESCRIPTION

[0076] Below is a combination of the embodiments and Figure 1-Figure 2 The present invention is further described in detail, but the embodiments of the present invention are not limited thereto.

[0077] The present invention discloses a heterogeneous information fusion method based on graph neural network, which specifically includes the following steps:

[0078] Step S1, collect information data, pre-process the information data, and use the pre-processed data as basic data.

[0079] Heterogeneous data usually comes from different data sources, and these data may have problems such as missing, duplicated, and noisy. Preprocessing the data means cleaning the data and identifying information data with basic problems such as duplication and missing.

[0080] Step S2, setting conversion standards for different categories of basic data, converting the basic data according to the conversion standards to obtain primary data.

[0081] Heterogeneous data may have different formats, structures, and units. Converting data requires unifying these different formats, structures, and units for subsequent analysis and processing. For example, converting location information into a unified geographic coordinate system.

[0082] Step S3, extract and identify corresponding entities based on the primary data and use them as standard entity objects.

[0083] When integrating heterogeneous data, it is necessary to identify and match the same entities or attributes in different data sources, such as user ID, mobile phone number, etc.

[0084] Step S4, obtaining the collection information of the primary data, judging the data quality of the primary data according to the collection information and the standard entity object, and obtaining the intermediate data according to the data quality matching.

[0085] Step S5, forming a heterogeneous information graph based on the medium data, selecting a graph neural network model to process the heterogeneous information graph, and obtaining a heterogeneous information fusion result.

[0086] Graph neural network refers to the use of neural networks to learn graph structure data, extract and discover features and patterns in graph structure data. Graph neural networks will use these heterogeneous relationships to integrate node information under different relationships through specific fusion mechanisms (such as attention mechanism, message passing mechanism, etc.) to generate richer node representations. Through the selected graph neural network, the fusion information results are further integrated through the graph neural network to obtain more accurate and subjective heterogeneous information fusion results.

[0087] In actual applications, after heterogeneous information fusion, it is usually necessary to further analyze and mine the fused data to obtain valuable information. However, if there are differences in data quality, the accuracy of analysis and mining will also be affected. For example, if there is noise or bias in the data, the results of the analysis may lead to misleading conclusions. Therefore, in the process of heterogeneous information fusion, information can be matched according to data quality, and poor quality data can be screened out to improve the effect and accuracy of heterogeneous information fusion.

[0088] The steps of obtaining the collection information of the primary data, judging the data quality of the primary data according to the collection information and the standard entity object, and obtaining the intermediate data according to the data quality matching are as follows:

[0089] Step S41, acquiring the collection information of the primary data, the collection information including the data generation time and collection related information.

[0090] The data generation time is the time point of data output. For example, if the collected data is user voice, the time point when the user speaks is the data generation time of the user voice.

[0091] Step S42, collecting related information including collection source, collection occasion and collection environment, and analyzing the accuracy of primary data based on the collection related information and standard entity objects.

[0092] Step S43, obtaining the effective duration of the primary data, and determining the validity of the primary data based on the data generation time and the effective duration.

[0093] Step S44, obtaining the data type of the primary data, and determining the completeness of the primary data in combination with the data generation time and the collection environment.

[0094] Step S45, respectively set the data weight ratios of accuracy, validity and completeness, and calculate the data quality according to the accuracy, validity, completeness and the corresponding data weight ratios.

[0095] Step S46, setting data quality standards, matching data that meet the data quality standards from the primary data and recording them as medium data.

[0096] For example, by setting a data quality threshold, it is considered that elementary data that reaches the data quality threshold can be matched with the standard entity object, and these data are recorded as medium data.

[0097] In actual application, the accuracy, validity and completeness of data will affect the quality of data. If the data is inaccurate, it will directly lead to wrong decisions and may even cause serious consequences. For example, if the standard entity object is a user, and the matched location data is inaccurate, the subsequent recommendation of user personnel and various business handling will be seriously affected. During the data entry process, if a field requires an email address, but the user enters a string of garbled characters or a string that does not conform to the email format, the data in this field is invalid. This invalid data will affect subsequent data processing and analysis. Similarly, incomplete data will also affect subsequent data analysis and affect the effect of information fusion.

[0098] The steps for analyzing the accuracy of primary data based on the collection of associated information and standard entity objects are as follows:

[0099] Step S421, determining whether the original outputter of the primary data has subjective consciousness based on the source of collection.

[0100] Subjective consciousness refers to the uniqueness and individuality of an individual when thinking, perceiving, and experiencing the world. It involves personal experience, emotion, and cognition, and is in contrast to objective facts and logical analysis. Therefore, if the original output is a person, then there is subjective consciousness, but if the original output is a machine device such as a sensor or detector, then there is no subjective consciousness. For example, if the source of the temperature of 25 degrees Celsius is a sensor, then the original output is the sensor, which does not have subjective consciousness.

[0101] Step S422: if the original outputter has no subjective consciousness, the historical output records of the original outputter are obtained, the data accuracy in the historical output records is extracted, and the data accuracy is used as the accuracy of the primary data.

[0102] If the original outputter has no subjective consciousness, the output data will not be affected by subjective factors. At this time, the accuracy of the primary data can be obtained based on the actual data accuracy. For example, the sensor detects the temperature and outputs the temperature data. The data error comes from the accuracy and failure of the sensor, but the sensor will not tamper with the temperature data subjectively.

[0103] Step S423, if the original outputter has subjective consciousness, the data output habit of the original outputter is obtained, and the data subjectivity of the primary data is determined according to the data output habit and the collection occasion.

[0104] Step S424, obtaining the emotional state of the original outputter according to the data output habit analysis, and obtaining the data emotional degree of the primary data according to the emotional state.

[0105] Step S425, obtaining the information fusion requirements of the standard entity object, and combining the data subjectivity and data sentiment to obtain the accuracy of the primary data.

[0106] In actual application, the accuracy of data is affected by the outputter. If the outputter has no subjective consciousness, the accuracy of the primary data is obtained according to the actual output accuracy of the outputter. If the outputter has subjective consciousness, the information given by the outputter will affect the accuracy of the data due to subjectivity and emotions. In the example process, the user represents the original outputter. For example, user A says that the temperature today is 26 degrees Celsius. Since user A has subjective consciousness, it is possible that user A feels that the weather is hot and says the temperature is higher, and the actual temperature is not 26 degrees Celsius. Therefore, when judging the accuracy of data, it is necessary to consider the subjective and emotional degree of the data and combine it with the standard entity object.

[0107] If the original outputter has subjective consciousness, the steps of obtaining the data output habits of the original outputter and determining the data subjectivity of the primary data according to the data output habits and the collection occasion are as follows:

[0108] Step S4231, collect words that describe subjective feelings and record them as subjective words.

[0109] Subjective consciousness refers to the uniqueness and individuality of an individual when thinking, perceiving, and experiencing the world. It involves personal experience, emotion, and cognition, and is in contrast to objective facts and logical analysis. For example, words such as "like", "feel", and "think".

[0110] Step S4232, determining whether the primary data is describing objective facts. If the primary data is describing objective facts, then counting the actual frequency AP of using subjective words in the primary data.

[0111] Objective facts can be verified by scientific methods or logical reasoning, and are not affected by personal subjective will. They exist independently in time and space. For example, "Today is October 25, 2024" is an objective fact because it is not affected by personal feelings or opinions, and can be judged by training models to determine whether it is an objective fact.

[0112] Step S4233, extracting the average frequency AB of subjective words used by the original outputter in the data output habit.

[0113] Step S4234, extracting the occurrence time AF of the objective fact described by the primary data, obtaining the contact time AJ of the original outputter's last contact with the objective fact associated information, and the contact frequency AL of the original outputter with the objective fact associated information.

[0114] Step S4235, collect historical collection data of collection occasions, and count the data accuracy rate AZ in the historical collection data.

[0115] The accuracy of information collected in different occasions is also different. For example, in daily situations, some wrong information may be collected due to users' jokes, and the data accuracy is lower. The higher the accuracy, the less erroneous content caused by subjectivity, and the lower the subjectivity. In more serious occasions, such as academic research, the accuracy of the collected data will be higher.

[0116] Step S4236, using the subjective correlation function to calculate the data subjective degree AG, the subjective correlation function is:

[0117]

[0118] Among them, a 1 、a 2 、a 3 is the scaling factor and is greater than 0, and AS is the data generation time.

[0119] Step S4237, if the elementary vocabulary is not describing objective facts, the data subjectivity is the maximum value.

[0120] If the original output is not describing objective facts, it is considered that the original output is describing personal emotions. For example, "I think this machine is very practical" is not describing objective facts, but only expressing the user's feelings. Therefore, the information is completely derived from the user's subjectivity, and the subjectivity is the maximum value.

[0121] In actual application, if the subjectivity contained in the information is greater, the less objective the information is, and the greater the subjective influence of the original outputter, thus reducing the accuracy of the information data. When some subjective words such as "think" and "feel" appear in the information, the information is mixed with the subjective judgment of the describer, reducing the credibility of the information. However, due to the different language habits of different users, some users are accustomed to adding subjective words in the language description, but in fact there is not much subjective content in it, so it is necessary to confirm it according to the average frequency of the user's daily use of subjective words. When the actual frequency is greater than the average frequency, it means that the user has added more subjective content, and the subjectivity of the information is greater. At the same time, when the user describes the objective facts, due to the different occurrence time of the objective facts, the user's memory is not deep, and some subjective content will be automatically used to fill in the objective facts. At this time, the user's subjective content exists in the information. The greater the time gap between the objective facts and the user's description, the more vague the memory. At the same time, if the user's last contact time is long, the memory is more vague, and the subjectivity is greater. The higher the contact frequency, the deeper the memory and the lower the subjectivity. For example, if user A is the original outputter and the output content is a description of the process of experiment A, experiment A took a long time to occur, and user A has not been exposed to the relevant information of experiment A, then the judgment data is highly subjective. User B often watches the operation video of experiment A, so the data of user B is less subjective.

[0122] The steps of obtaining the emotional state of the original outputter according to the data output habit analysis and obtaining the data emotional degree of the primary data according to the emotional state are as follows:

[0123] Step S4241, obtaining the emotion fluctuation degree of the original outputter, calculating the average emotion fluctuation degree according to the emotion fluctuation degree and recording it as the original emotion degree BY.

[0124] The degree of emotional fluctuation of the original outputter can be determined through physiological information such as the heart rate of the original outputter, or authorized information such as facial expressions. For example, the average difference between the heart rate and the normal average heart rate is calculated to obtain the average degree of emotional fluctuation.

[0125] Step S4242, record the words, symbols and expressions that express emotions as emotion words, and count the actual usage frequency of the emotion words in the primary data.

[0126] Some words are used to express the emotions of the original outputter, such as "awkward", "bored", "anxious", etc.

[0127] Step S4243, extracting the average frequency of emotion words used by the original outputter in the data output habit, calculating the difference between the actual frequency of use of the emotion words and the average frequency of use of the emotion words and recording it as the actual emotion degree BS.

[0128] Some users like to use emotional words in their daily lives, so using emotional words does not necessarily mean that the user has emotions. For example, on the Internet, users use smiley faces to express friendliness, but in fact the user does not have any emotions, they just add smiley faces out of habit. When a user outputs "hahahahaha", it is just a simple message response, while outputting "hahahahahahahahahahahahahahahahaha" expresses happiness. Therefore, the actual emotional level of the user can be judged based on the frequency of use of emotional words.

[0129] Step S4244, extract the objective facts of the original outputter in the primary data, and search for the fact type of the objective facts.

[0130] Extract the objective facts given by the original outputter and search for objective facts of the same type. For example, if the objective fact described by the original outputter is that today's weather is sunny, then the fact type of the objective fact is weather.

[0131] Step S4245, searching for the output emotion fluctuation degree of the information outputter of the same type of information of the fact type, calculating the average value of the output emotion fluctuation degree and recording it as the information emotion degree BX.

[0132] The actual emotion degree method is used to obtain the degree of output emotion fluctuation of the information outputter, and then the average value of the output emotion fluctuation degree of all information outputters is calculated to obtain the information emotion degree.

[0133] In addition to being affected by the person, the emotions in the information data are also affected by the objective facts being described. For example, when describing an objective fact whose fact type is a date, most outputters will be relatively calm. However, when describing an objective fact whose fact type is the loss of a valuable item, most outputters will be anxious and have greater emotional fluctuations. If describing the information generally causes emotional fluctuations in the outputter, the information will carry more emotions.

[0134] Step S4246, using the emotion correlation function to calculate the data emotion degree BQ, emotion correlation function:

[0135]

[0136] Among them, b 1 、b 2 、b 3 is the scaling factor and is greater than 0.

[0137] In actual use, in addition to the original outputter's subjective influence on the accuracy of information, the original outputter's emotions will also affect the accuracy of the data. Emotional state affects the individual's cognitive process and information processing method. When the emotions are excited, the information will be biased, thus affecting the accuracy of the information. When the objective facts occur at a longer time, the user's emotions will tend to be calmer, so the emotions carried in the description process will be less. If the user's original emotional degree is greater, it means that the user's emotions are easily affected, so it is easier to carry personal emotions in the information output process.

[0138] The steps to obtain the information fusion requirements of standard entity objects and obtain the accuracy of primary data by combining data subjectivity and data sentiment are as follows:

[0139] Step S4251, obtaining the information fusion requirements of the standard entity object, and determining whether the information fusion requirements only require objective facts.

[0140] Different industries have different requirements for heterogeneous information fusion. For example, in the field of intelligent manufacturing, heterogeneous system integration technology can break through the barriers between different devices and systems and achieve seamless connection and flow of data. In this case, heterogeneous information fusion only requires objective facts, that is, accurate data. However, in the field of product recommendation, heterogeneous information fusion needs to obtain the actual feelings of users. The subjective feelings and emotions of users play a vital role in product recommendations, so subjective data and emotional data are needed.

[0141] Step S4252: If the information fusion requires only objective facts, a data subjectivity threshold is set, and primary data that does not reach the data subjectivity threshold is screened and recorded as the first data. The data emotion of the first data is used as the accuracy of the primary data, and the accuracy of the primary data other than the first data is 0.

[0142] Due to the influence of subjective factors, the supplemented content may deviate from or be wrong from the actual events. If only objective facts are needed and subjective content does not represent objective facts, then the data with greater subjectivity will be filtered out to reduce the information data of non-objective facts. Emotions will also affect the accuracy of objective descriptions, so after filtering according to the subjective degree of the data, the emotional degree of the data is used as the accuracy of the primary data.

[0143] Step S4253: If the information fusion requirement is not just objective facts, the first weight ratios of the data subjectivity and the data emotion are set respectively, and the accuracy of the elementary data is calculated based on the data subjectivity, the data emotion and the corresponding first weight ratio.

[0144] In actual application, the accuracy of data needs to be evaluated according to the actual situation. For information fusion that only requires objective facts, the subjectivity and emotions of the original outputter become interference items of the information, so the accuracy of information mixed with subjectivity and emotions is lower. However, for information fusion that requires the subjectivity and emotions of the original outputter, information mixed with subjectivity and emotions can provide more valuable information content, so the accuracy is higher.

[0145] The steps of obtaining the effective duration of the primary data and determining the effectiveness of the primary data according to the data generation time and the effective duration are as follows:

[0146] Step S431, obtain the effective duration of the elementary data, obtain the real time, and calculate the time difference between the data generation time and the real time.

[0147] Step S432, determine whether the time difference exceeds the valid time length. If the time difference exceeds the valid time length, the validity of the primary data is 0.

[0148] If the data has exceeded the validity period, it means that the data is invalid and useless, so the validity is 0.

[0149] Step S433: if the time difference does not exceed the effective duration, the elementary data of the same fact type are formed into a type data set.

[0150] Among the collected data, a lot of data express the same fact. For example, a user enters address information on platform 1 and also enters address information on platform 2. Both pieces of information describe the user's address, and the fact type is the same.

[0151] Step S434, extracting the actual meaning of the data in the type data set, and extracting the actual meaning with the largest number as the standard meaning.

[0152] The same type of data describes an entity or fact, but may also be different. For example, user A is called "Zhang San" on platform 1, "Zhang San" on platform 2, and "Mr. Zhang" on platform 3. Therefore, the actual meaning of the largest number is that the user's name is "Zhang San". "Zhang San" is the standard meaning of the user's name.

[0153] Step S435, extracting the actual meaning of the elementary data, comparing and obtaining the similarity between the actual meaning and the standard meaning and recording it as the validity of the elementary data.

[0154] In actual application, the greater the deviation of the data, the lower the validity of the data. For example, if user A is called "Sunflowers are Singing" on the No. 4 platform, the similarity with "Zhang San" is too low. Obviously, the deviation from the user's name is too large and it cannot provide valid information for the user's name, so the validity is low.

[0155] The steps to obtain the data type of the primary data and determine the integrity of the primary data based on the data generation time and collection environment are as follows:

[0156] Step S441, obtaining the necessary attributes of the primary data, and judging whether the primary data is complete according to the necessary attributes. If the primary data is incomplete, the completeness of the primary data is 0.

[0157] Different data has its necessary attributes, and data with necessary attributes is considered complete. For example, for text data, symbols such as periods, exclamation marks, and question marks indicate that the sentence is complete. For temperature acquisition equipment, the data is considered complete only if it ends in degrees Celsius.

[0158] Step S442, if the primary data is complete, the data type of the primary data is obtained, the basic completeness standard of the data type is extracted, and the proportion of the primary data that meets the basic completeness standard is obtained and recorded as the completeness proportion.

[0159] The required attributes only indicate the basic completeness of the data, which is not necessarily complete for information fusion. For example, address information can be divided into different minimum units from large to small. It can be divided into districts, buildings, and house numbers. According to the data type of the primary data, the basic completeness standard generally required for the information fusion is obtained. The completeness ratio is obtained based on the content ratio of the basic completeness standard data contained in the primary data. For example, the basic completeness standard is province, city, district, town, and building, and the information ends at the district, then the completeness ratio is 60%.

[0160] Step S443, judging the interference degree caused by the acquisition environment to the primary data according to the acquisition environment.

[0161] Step S444, respectively set the integrity ratio of the integrity proportion and the interference degree, and calculate the integrity of the primary data according to the integrity proportion, the interference degree and the corresponding integrity weight ratio.

[0162] In actual applications, incomplete data may lead to incomplete information, thus affecting the effect of information fusion and further affecting the use of information after fusion. For example, in a prediction model, if the driver's complete information (such as age and gender) is missing, the prediction results of the model may be seriously affected.

[0163] The steps to determine the degree of interference caused by the collection environment to the primary data are as follows:

[0164] Step S4431, obtain the interference factors that affect the elementary data, count the interference factors contained in the collection environment and record them as interference factors.

[0165] Interference factors are factors that interfere with primary data. For example, when collecting user voice, noise will interfere with the user voice, so noise is an interference factor. If there is no noise in the collection environment, then noise is not an interference factor. If there is noise, then noise is an interference factor.

[0166] Step S4432, obtaining the interference range of the interference factor, obtaining the actual interference value of the interference factor, and obtaining the degree of the interference factor according to the actual interference value and the interference range.

[0167] The interference range refers to the range of interference levels that can be caused by interference factors. For example, the interference level of noise ranges from 50 decibels to 100 decibels, and the actual interference value is 80 decibels, then (80-50) / (100-50)=60%, so the interference factor level is 60%.

[0168] Step S4433, obtaining the basic interference degree caused by the acquisition environment according to the sum of all interference factors and the corresponding interference factor degrees.

[0169] The values ​​of all interference factors are superimposed and calculated to obtain the basic interference level.

[0170] Step S4434, counting the transmission duration and transmission times of the primary data, and obtaining the transmission interference degree according to the transmission duration and transmission times.

[0171] Step S4435, the basic interference degree and the transmission interference degree are superimposed and calculated to obtain the actual interference degree of the primary data.

[0172] In actual use, when data needs to be transmitted multiple times, there is a certain risk of data loss in each transmission. For example, in UDP transmission, if there is no appropriate retransmission mechanism, the data packet may be lost due to network instability and other reasons. As the number of transmissions increases, the cumulative risk of data loss will also increase. Long-term transmission may increase the risk of data loss. For example, in the case of unstable network, long-term transmission is more susceptible to network interruption or data packet loss. Data loss will seriously affect the integrity of the data, and data preprocessing can only clean up simple missing data. Some data is lost during transmission and may not be discovered. Therefore, the longer the transmission time and the more transmission times, the lower the data integrity.

[0173] A heterogeneous information fusion system based on graph neural network, by applying the above-mentioned heterogeneous information fusion method based on graph neural network, includes:

[0174] The data processing module collects information data, pre-processes the information data, and uses the pre-processed data as basic data.

[0175] The conversion standard module sets the conversion standards for different categories of basic data, and converts the basic data according to the conversion standards to obtain primary data.

[0176] The entity recognition module extracts and recognizes the corresponding entities based on the primary data and uses them as standard entity objects.

[0177] The data quality module obtains the collection information of the primary data, judges the data quality of the primary data according to the collection information and the standard entity object, and obtains the intermediate data according to the data quality matching.

[0178] The fusion result module forms a heterogeneous information graph based on the medium data, selects the graph neural network model to process the heterogeneous information graph, and obtains the heterogeneous information fusion result.

[0179] The implementation principle of this system is: first, collect information data through the data processing module, pre-process the information data, and use the pre-processed data as basic data. The conversion standard module sets the conversion standards for different categories of basic data, converts the basic data according to the conversion standards, and obtains primary data. The entity recognition module extracts and recognizes the corresponding entity based on the primary data and uses it as a standard entity object. The data quality module obtains the collection information of the primary data, which includes the data generation time and collection related information. The collection related information includes the collection source, collection occasion and collection environment. The accuracy of the primary data is analyzed according to the collection related information and the standard entity object. The validity of the primary data is obtained by judging the data generation time and the effective duration, and the completeness of the primary data is obtained by combining the data generation time and the collection environment. The data quality is calculated by comprehensively calculating the accuracy, validity and completeness, and the medium data is obtained according to the data quality matching. Finally, the fusion result module forms a heterogeneous information graph based on the medium data, selects the graph neural network model to process the heterogeneous information graph, and obtains the heterogeneous information fusion result.

[0180] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A heterogeneous information fusion method based on graph neural network, characterized in that: The following steps are involved: Collect information data, pre-process the information data, and use the pre-processed data as basic data; Set conversion standards for different categories of basic data, convert the basic data according to the conversion standards, and obtain primary data; According to the elementary data extraction, the corresponding entities are identified and used as standard entity objects; Obtaining the collection information of primary data, judging the data quality of the primary data according to the collection information and standard entity objects, and obtaining the intermediate data according to the data quality matching; A heterogeneous information graph is formed according to the medium data, and a graph neural network model is selected to process the heterogeneous information graph to obtain the heterogeneous information fusion result.

2. According to claim 1, a heterogeneous information fusion method based on graph neural network is characterized in that: The steps of obtaining the collection information of the primary data, judging the data quality of the primary data according to the collection information and the standard entity object, and obtaining the intermediate data according to the data quality matching are specifically as follows: Acquire collection information of primary data, wherein the collection information includes data generation time and collection related information; The collection-related information includes the collection source, collection occasion and collection environment, and the accuracy of the primary data is analyzed based on the collection-related information and the standard entity object; Obtain the effective duration of the primary data, and determine the validity of the primary data based on the data generation time and effective duration; Obtain the data type of the primary data, and determine the completeness of the primary data based on the data generation time and collection environment; The data weight ratios of accuracy, validity and completeness are set respectively, and the data quality is calculated according to the accuracy, validity, completeness and the corresponding data weight ratios; Set data quality standards, match data that meet the data quality standards from primary data and record them as medium data.

3. According to claim 2, a heterogeneous information fusion method based on graph neural network is characterized in that: The step of analyzing the accuracy of the primary data based on the collected association information and the standard entity object is specifically as follows: According to the source of the collection, the original outputter of the primary data is determined to determine whether the original outputter has subjective consciousness; If the original outputter has no subjective consciousness, then the historical output records of the original outputter are obtained, the data accuracy in the historical output records is extracted, and the data accuracy is used as the accuracy of the primary data; If the original outputter has subjective consciousness, then the data output habits of the original outputter are obtained, and the data subjectivity of the primary data is determined based on the data output habits and the collection occasions; The emotional state of the original outputter is obtained by analyzing the data output habits, and the data emotional degree of the primary data is obtained according to the emotional state; Obtain the information fusion requirements of standard entity objects, and combine the data subjectivity and data sentiment to obtain the accuracy of primary data.

4. According to claim 3, a heterogeneous information fusion method based on graph neural network is characterized in that: If the original outputter has subjective consciousness, the steps of obtaining the data output habit of the original outputter and determining the data subjectivity of the primary data according to the data output habit and the collection occasion are specifically as follows: Collect words that describe subjective feelings and record them as subjective words; Determine whether the primary data is describing objective facts. If the primary data is describing objective facts, then count the actual frequency AP of subjective words used in the primary data. Extract the average frequency AB of subjective words used by the original outputter in the output habits of data; Extract the occurrence time AF of the objective fact described by the primary data, obtain the contact time AJ of the original outputter's last contact with the objective fact-related information, and the contact frequency AL of the original outputter with the objective fact-related information; Collect historical data collected at the collection site, and count the data accuracy rate AZ in the historical data collected; The subjective degree correlation function is used to calculate the data subjective degree AG, subjective degree correlation function: Among them, a1, a2, a3 are scale factors and are greater than 0, and AS is the data generation time; If the elementary vocabulary is not describing objective facts, the subjectivity of the data is at its maximum.

5. The heterogeneous information fusion method based on graph neural network according to claim 4 is characterized in that: The step of obtaining the emotional state of the original outputter according to the data output habit analysis and obtaining the data emotional degree of the primary data according to the emotional state is specifically as follows: The emotional fluctuation degree of the original outputter is obtained, and the average emotional fluctuation degree is calculated according to the emotional fluctuation degree and recorded as the original emotional degree BY; The words, symbols and expressions that express emotions are recorded as emotional words, and the actual usage frequency of emotional words in the elementary data is counted; Extract the average frequency of emotional words used by the original outputter in the data output habits, calculate the difference between the actual frequency of emotional words used and the average frequency of emotional words used and record it as the actual emotional degree BS; Extract objective facts of the original output in the primary data and find the fact type of the objective facts; Find the output emotion fluctuation degree of the information outputter of the same type of information of the fact type, calculate the average value of the output emotion fluctuation degree and record it as the information emotion degree BX; The emotional degree correlation function is used to calculate the data emotional degree BQ, emotional degree correlation function: Among them, b1, b2, and b3 are proportional factors and are greater than 0.

6. The heterogeneous information fusion method based on graph neural network according to claim 5 is characterized in that: The step of obtaining the information fusion requirements of the standard entity object and obtaining the accuracy of the primary data by combining the data subjectivity and the data sentiment is specifically as follows: Obtain the information fusion requirements of standard entity objects and determine whether the information fusion requirements only require objective facts; If the information fusion requirement is only objective facts, then a data subjectivity threshold is set, and the primary data that does not reach the data subjectivity threshold is screened and recorded as the first data. The data sentiment of the first data is used as the accuracy of the primary data, and the accuracy of the primary data other than the first data is 0; If the information fusion requirement is not just objective facts, the first weight ratios of the data subjectivity and the data sentiment are set respectively, and the accuracy of the elementary data is calculated based on the data subjectivity, the data sentiment and the corresponding first weight ratios.

7. The heterogeneous information fusion method based on graph neural network according to claim 6 is characterized in that: The step of obtaining the effective duration of the primary data and determining the effectiveness of the primary data according to the data generation time and the effective duration is specifically as follows: Get the effective duration of the elementary data, get the real time, and calculate the time difference between the data generation time and the real time; Determine whether the time difference exceeds the valid duration. If the time difference exceeds the valid duration, the validity of the primary data is 0; If the time difference does not exceed the effective duration, the elementary data of the same fact type are formed into a type data set; Extract the actual meaning of the data in the type data set, and extract the actual meaning with the largest number as the standard meaning; The actual meaning of the primary data is extracted, and the similarity between the actual meaning and the standard meaning is compared and recorded as the validity of the primary data.

8. The heterogeneous information fusion method based on graph neural network according to claim 7 is characterized in that: The steps of obtaining the data type of the elementary data and determining the integrity of the elementary data in combination with the data generation time and the collection environment are specifically as follows: Obtain the necessary attributes of the primary data, and determine whether the primary data is complete based on the necessary attributes. If the primary data is incomplete, the completeness of the primary data is 0; If the primary data is complete, the data type of the primary data is obtained, the basic completeness standard of the data type is extracted, and the proportion of the primary data that meets the basic completeness standard is obtained and recorded as the completeness proportion; Determine the degree of interference caused by the collection environment to the primary data based on the collection environment; The completeness ratio of the completeness proportion and the completeness weight ratio of the interference degree are set respectively, and the completeness of the primary data is calculated according to the completeness ratio, the interference degree and the corresponding completeness weight ratio.

9. The heterogeneous information fusion method based on graph neural network according to claim 8 is characterized in that: The step of judging the interference degree of the collection environment to the primary data according to the collection environment is specifically as follows: Obtain the interference factors that affect the primary data, count the interference factors contained in the collection environment and record them as interference factors; Obtain the interference range of the interference factor, obtain the actual interference value of the interference factor, and obtain the degree of the interference factor according to the actual interference value and the interference range; The basic interference degree caused by the collection environment is obtained by summing up all interference factors and the corresponding interference factor degrees; Count the transmission time and number of transmissions of the elementary data, and obtain the degree of transmission interference according to the transmission time and number of transmissions; The basic interference degree and the transmission interference degree are superimposed and calculated to obtain the actual interference degree of the primary data.

10. A heterogeneous information fusion system based on graph neural network, characterized in that: By applying a heterogeneous information fusion method based on a graph neural network as described in any one of claims 1 to 9, comprising: The data processing module collects information data, pre-processes the information data, and uses the pre-processed data as basic data; The conversion standard module sets the conversion standards for different categories of basic data, and converts the basic data according to the conversion standards to obtain primary data; The entity recognition module extracts and recognizes the corresponding entity based on the primary data and uses it as a standard entity object; The data quality module obtains the collection information of the primary data, determines the data quality of the primary data based on the collection information and the standard entity object, and obtains the intermediate data based on the data quality matching; The fusion result module forms a heterogeneous information graph based on the medium data, selects the graph neural network model to process the heterogeneous information graph, and obtains the heterogeneous information fusion result.

Images