CN104850632A - Generic similarity calculation method and system based on heterogeneous information network - Google Patents

Abstract

The invention discloses a general similarity calculation method based on a heterogeneous information network, comprising: step 1, preprocessing the input data set to ensure the validity of the input data; step 2, extracting metadata to extract the input The description information of the data, and store the description information in the metadata database; step 3, establish the heterogeneous information network model through user interaction, and store the network model; step 4, adopt the similarity calculation method based on the meta-path in the heterogeneous information network Carry out similarity calculation; step 5, similarity post-processing to form a total similarity as the final output. Beneficial effects of the present invention: adopting heterogeneous information network modeling, a general similarity calculation method is proposed; different types of data sets can be processed; various similarity calculation requirements can be met; users can specify multiple calculation methods and The result post-processing method has a high degree of freedom of choice, improves the calculation accuracy and efficiency, and better solves the problem of information overload.

Description

A general method and system for similarity calculation based on heterogeneous information network

technical field

The present invention relates to the fields of information technology and Internet technology, in particular, to a general heterogeneous information network-based similarity calculation method and system.

Background technique

With the development of information technology and the Internet, people have gradually entered the era of information overload from the early data scarcity. Especially in the current era of big data, how to solve the problem of information overload and extract valuable information from massive data is a key issue that people urgently need to solve. Whether in various information retrieval systems or in various personalized recommendation systems and applications, information similarity calculation is a key technology in these systems and applications, and usually plays a decisive role in the processing accuracy of related systems and applications.

Heterogeneous information network is a relatively new research field emerging with the development of social network, also known as heterogeneous social network or multi-relational social network. The heterogeneous information network specifies the type constraints on object collections and the relationship constraints between objects through network patterns. These constraints make the heterogeneous information network semi-structured, thus guiding people to better explore network semantics. Heterogeneous information networks can be constructed from many interconnected social, scientific, engineering, and large-scale data sets for commercial applications, as well as e-commerce such as Amazon and eBay, online movie databases such as IMDb (Internet MovieDatabase), and various database applications.

As a general big data mining tool, the heterogeneous information network has a good expressive power for processing the relationship and structural features between data. Modeling the relationship in the real world through the heterogeneous information network can effectively Calculate the similarity between information in the real world.

At present, some traditional calculation methods for calculating the similarity between entities in the real world are usually only for specific data, the method is simple and fixed, and cannot well reflect the rich relationship between entities in the real world, lacking a general calculation method and The framework usually uses some simple similarity calculation methods, and only considers few relatively fixed attributes between data and data in the calculation process. When considering different types of data and different attributes of data, it must be re- Consider the calculation method, and the system needs to be re-modified, resulting in poor versatility, low calculation efficiency and low accuracy of the results, which cannot meet the requirements of today's big data era.

Contents of the invention

To solve the above problems, the object of the present invention is to provide a general similarity calculation method and system that can select corresponding data attributes and calculation methods according to user requirements.

The present invention provides a general similarity calculation method based on heterogeneous information network, including:

Step 1, preprocessing the input data set to ensure the validity of the input data;

Step 2, perform metadata extraction, extract the description information of the input data, and store the description information in the metadata database, wherein the description information includes global information of the overall situation of the input data set, each item The local information of the record and the conversion and corresponding information between the identifier of the data attribute and the internal representation;

Step 3, the user selects the entities and data attributes involved in the similarity calculation, queries the corresponding metadata, displays the data type and value range of each metadata, and prompts the user to select the metadata according to the preset processing library The processing method, with the entity as the central node, each entity is linked according to the semantic relationship, each attribute is connected to the corresponding entity, a heterogeneous information network model is generated, and the heterogeneous information network model is stored;

Step 4, in the heterogeneous information network mode, after the user specifies the data attribute and meta-path, perform similarity calculation according to the selected processing method;

In step 5, according to the weight of the selected data attribute, the multiple similarity results calculated in step 4 are merged to obtain a unified similarity result, and the format of the unified similarity result is converted and output to the user.

As a further improvement of the present invention, in step 1, the preprocessing of the input data set includes data cleaning and data integration;

in,

Data cleaning is to perform format conversion, eliminate noise, delete inconsistent data, perform data cleaning on the input data set, remove useless noise data, and perform corresponding format conversion;

Data integration is combining data from multiple data sources.

As a further improvement of the present invention, step 4 specifically includes:

Step 401, the user specifies data attributes and selects a meta path, if multiple data attributes are involved, the user specifies the link sequence of multiple data attributes to form a meta path;

Step 402, after the user specifies a single data attribute, retrieve the data in the corresponding data attribute, and construct the data into an adjacency matrix, if the user specifies a meta-path, construct the adjacency matrix with all the relationships on the meta-path , and then standardize the adjacency matrix;

In step 403, after the adjacency matrix is normalized, a matrix operation is used to obtain a similarity result matrix of the corresponding attribute.

As a further improvement of the present invention, the format of the input data set in step 1 includes a relational database format, a NOSQL database format, an ARFF file, a CSV file, a text file and an excel spreadsheet.

As a further improvement of the present invention, in step 2, the global information includes the number of records in the input data set and the number of data attributes in each record, and the local information includes the identifier of each data attribute, data type and value range.

The present invention also provides a general similarity calculation system based on heterogeneous information networks, including:

The processing module preprocesses the input data set to ensure the validity of the input data;

The extraction module extracts metadata, extracts the description information of the input data, and stores the description information in the metadata database, wherein the description information includes global information of the overall situation of the input data set, each item The local information of the record and the conversion and corresponding information between the identifier of the data attribute and the internal representation;

In the modeling module, the user selects the entities and data attributes involved in the similarity calculation, queries the corresponding metadata, displays the data type and value range of each metadata, and prompts the user to select the metadata according to the preset processing library. The data processing method, with the entity as the central node, each entity is linked according to the semantic relationship, each attribute is connected to the corresponding entity, a heterogeneous information network model is generated, and the heterogeneous information network model is stored;

Calculation module, in the heterogeneous information network mode, after the user specifies the data attribute and the meta path, calculate the similarity according to the selected processing method;

The post-processing module fuses the multiple similarity results calculated in step 4 according to the weights of the selected data attributes to obtain a unified similarity result, converts the format of the unified similarity result, and outputs it to the user.

As a further improvement of the present invention, the preprocessing of the input data set by the processing module includes data cleaning and data integration;

in,

Data cleaning is to perform format conversion, eliminate noise, delete inconsistent data, perform data cleaning on the input data set, remove useless noise data, and perform corresponding format conversion;

Data integration is combining data from multiple data sources.

As a further improvement of the present invention, the calculation module includes:

Specify the module, the user specifies the data attribute, and selects the meta path. If multiple data attributes are involved, the user specifies the link sequence of multiple data attributes to form the meta path;

The construction module, after the user specifies a single data attribute, calls the data in the corresponding data attribute, and constructs these data into an adjacency matrix, if the user specifies a meta-path, then constructs the adjacency matrix with all the relationships on this meta-path , and then standardize the adjacency matrix;

The operation module, after the adjacency matrix is normalized, uses matrix operation to obtain the similarity result matrix of the corresponding attribute.

As a further improvement of the present invention, the format of the input data set in the processing module includes a relational database format, a NOSQL database format, an ARFF file, a CSV file, a text file and an excel spreadsheet.

As a further improvement of the present invention, in the extraction module, the global information includes the number of records in the input data set and the number of data attributes in each record, and the local information includes an identifier for each data attribute , data type and value range.

The beneficial effects of the present invention are:

1. Using heterogeneous information network modeling, it better solves the problem of similarity calculation that is becoming more and more important at present, and has broad application prospects and use value;

2. It has good versatility and compatibility, and can handle different types of data sets;

3. It can meet the needs of users for multiple similarity calculations of data objects. It can be used to calculate the similarity between objects of the same type, and can also calculate the similarity between different objects. It is suitable for practical applications and systems in various fields . For example, the most frequently used calculations in the recommendation system are between users and users, between items and items, and between users and items, and can also calculate the similarity between system objects in the search system;

4. The user can specify a variety of similarity calculation methods and result post-processing methods, which are flexible and convenient, with a high degree of freedom for users to choose, and significantly improve the accuracy of calculation. At the same time, it has obvious advantages in calculation efficiency, which is a good solution Information overload problem.

Description of drawings

Fig. 1 is a flow chart of a general similarity calculation method based on a heterogeneous information network according to an embodiment of the present invention;

Fig. 2 is a flow chart of step 4 in a general heterogeneous information network-based similarity calculation method described in an embodiment of the present invention;

3 is a schematic diagram of a heterogeneous information network mode in a general heterogeneous information network-based similarity calculation method according to an embodiment of the present invention;

4 is a system diagram of a general-purpose similarity calculation system based on a heterogeneous information network according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of a calculation module in a general heterogeneous information network-based similarity calculation system according to an embodiment of the present invention.

Detailed ways

The present invention will be described in further detail below through specific embodiments and in conjunction with the accompanying drawings.

Embodiment 1, as shown in FIG. 1, a general similarity calculation method based on a heterogeneous information network according to an embodiment of the present invention includes:

Step 1, preprocessing the input data set to ensure the validity of the input data;

The following is an example of the movie recommendation dataset MovieLens100k provided by the University of Minnesota. The specific implementation is as follows: In the case of only using movie recommendations, a large amount of user data is redundant and needs to be removed. In this data set, there is a lack of information such as movie actors (Actor), director (Director), and the data set provides a link from the movie to the Internet movie data set IMDb. Combine the data set MovieLens 100K and the data set IMDb to obtain effective data.

Step 2, perform metadata extraction, extract the description information of the input data, and store the description information in the metadata database, where the description information includes the global information of the overall situation of the input data set, the local information of each record, and the attributes of the data Conversion and corresponding information between identifiers and internal representations;

In the MovieLens 100K data set, the global information includes: there are 1682 movies, a total of 943 users and 100,000 movie ratings, and the ratings range from 1 to 5. Partial information includes: identifiers are actor (Actor), release time (Year), director (Director), movie type (Genre), country (Country) and keyword (Keyword); partial information such as actor attributes, data type is character String, with a value range of 20 characters, requires an exact match. Regarding the conversion of data attribute identifiers and internal representations, the internal attributes actor (Actor) in MovieLens100K is marked as attribute 1 (A1), the release time (Year) is marked as attribute 2 (A2), and the director (Director) is marked as attribute 3 ( A3), the movie type (Genre) is marked as attribute 4 (A4), the country (Country) is marked as attribute 5 (A5), and the keyword (Keyword) is marked as attribute 6 (A6), so that the data as a whole can be considered more, and it is not necessary to Consider the specific situation of each data attribute for easy processing. Extract this information and save it for subsequent work calls.

Step 3, the user selects the entities and data attributes involved in the similarity calculation, queries the corresponding metadata, displays the data type and value range of each metadata, and prompts the user to select the metadata processing according to the preset processing library The method takes the entity as the central node, links each entity according to the semantic relationship, and connects each attribute to the corresponding entity, generates a heterogeneous information network model, and stores the heterogeneous information network model;

In the MovieLens 100K dataset, to further improve the commonly used data attributes of movies, it is necessary to further expand the movie attributes from the IMDb database. The selected entity is a movie, and the selected data attributes are: actor (Actor), release time (Year), director (Director), movie type (Genre), country (Country), keyword (Keyword), then the heterogeneous information network of the movie The pattern is shown in Figure 3.

Step 4, in the heterogeneous information network mode, after the user specifies the data attribute and meta-path, the similarity calculation is performed according to the selected processing method.

As shown in Figure 2, it specifically includes the following steps:

Step 401, the user specifies data attributes and selects a meta path, if multiple data attributes are involved, the user specifies the link sequence of multiple data attributes to form a meta path;

The similarity of heterogeneous information networks is usually determined by data attributes, so the user needs to specify which data attribute is used to calculate the similarity. Movies with the same actors are considered similar if popular actors are used to guarantee the movie box office.

Step 402, after the user specifies a single data attribute, retrieve the data in the corresponding data attribute, and construct these data into an adjacency matrix, if the user specifies a meta-path, construct an adjacency matrix for all relationships on this meta-path, and then Normalize the adjacency matrix;

In the MovieLens 100K data set, select the data attribute as actor (Actor), and you can get the adjacency matrix of actor attributes between movies. This similarity is based on a selected meta-path, that is, for a movie (Movie) , you can explore the path from the movie (Movie) to the actor (Actor) and then to the movie, such as the following path: Movie-(Actor-Movie-Actor) ^r -Movie; where r is a non-negative integer.

Step 403, after the adjacency matrix is normalized, use the matrix operation to obtain the similarity result matrix of the corresponding attribute;

If the meta-path is P(A ₁ A ₂ ... A _i ... A _l ), where P is the meta-path, A _i represents the attribute of the data, and R _i is defined as the relationship between A _i and A _i+1 , that is, there is an edge between A _i and A _i+1 . Then the similarity calculation formula between arbitrary objects s and t (s is the object of attribute _A1 , _t is the object of attribute A1) can be obtained.

HeteSim similarity can be used for calculation, the formula is:

The functions O(R) and I(R) represent the out-degree and in-degree of the relation R, respectively.

It can also be calculated using PathSim similarity, the formula is:

$\mathrm{<span\; class="notranslate">\; P</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; h</span>}\mathrm{<span\; class="notranslate">\; S</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; \&Right\; Arrow;</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; :</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; \&Right\; Arrow;</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; |</span>}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; \&Right\; Arrow;</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; :</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; \&Right\; Arrow;</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; \&Right\; Arrow;</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; :</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; \&Right\; Arrow;</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

Among them, p _s→t , p _s→s , and p _t→t are path instances between s and t, s and s, and t and t, respectively.

Step 5, according to the weight of the selected data attribute, the multiple similarity results calculated in step 4 are merged to obtain a unified similarity result, and the format of the unified similarity result is converted and output to the user;

According to user needs, the weight can take various methods such as the average value, maximum value, minimum value, and weighted average of the data attribute.

Further, the format of the input data set in step 1 includes a relational database form, a NOSQL database form, an ARFF file, a CSV file, a text file and an excel spreadsheet.

Further, in step 2, the global information includes the number of records in the input data set and the number of data attributes in each record, and the local information includes the identifier, data type and value range of each data attribute.

The pre-set processing library in step 3 is the basic processing method library for data similarity calculation. Generally, the types of data involved are limited to several types, such as numerical value, time, string, etc. Different processing needs to be done according to the situation when performing similarity calculation. If the attribute is a numeric data type, you can set them to be strictly equal, equal by a certain percentage or numerical error; if the attribute is a time data type, you can consider the same day, the same month, the same year to be considered equal, or a certain percentage difference or The value range is equal; if the attribute is a string data type, it can usually be set to be strict equality, approximate matching, etc. These commonly used processing methods are general solutions provided by this method for users to choose during use. Users can also define their own custom matching methods, for example, through user interaction. For example, the user can only limit the time to a certain day.

Embodiment 2, as shown in Figure 4, the present invention also provides a general similarity calculation system based on heterogeneous information networks, including:

The processing module preprocesses the input data set to ensure the validity of the input data;

The extraction module extracts metadata, extracts the description information of the input data, and stores the description information in the metadata database, where the description information includes the global information of the overall situation of the input data set, the local information of each record, and the attributes of the data Conversion and corresponding information between identifiers and internal representations;

In the modeling module, the user selects the entities and data attributes involved in the similarity calculation, queries the corresponding metadata, displays the data type and value range of each metadata, and prompts the user to select the metadata according to the preset processing library. The processing method takes the entity as the central node, links each entity according to the semantic relationship, and connects each attribute to the corresponding entity, generates a heterogeneous information network model, and stores the heterogeneous information network model;

Calculation module, in the heterogeneous information network mode, after the user specifies the data attribute and meta path, the similarity calculation is performed according to the selected processing method;

The post-processing module fuses the multiple similarity results calculated in step 4 according to the weights of the selected data attributes to obtain a unified similarity result, converts the format of the unified similarity result, and outputs it to the user.

Further, the preprocessing of the input data set by the processing module includes data cleaning and data integration;

in,

Data cleaning is to perform format conversion, eliminate noise, delete inconsistent data, perform data cleaning on the input data set, remove useless noise data, and perform corresponding format conversion;

Data integration is combining data from multiple data sources.

As shown in Figure 5, the calculation module includes:

Specify the module, the user specifies the data attribute, and selects the meta path. If multiple data attributes are involved, the user specifies the link sequence of multiple data attributes to form the meta path;

The construction module, after the user specifies a single data attribute, retrieves the data in the corresponding data attribute and constructs the data into an adjacency matrix. If the user specifies a meta path, all relationships on the meta path are constructed into an adjacency matrix, and then Normalize the adjacency matrix;

In the operation module, after the adjacency matrix is normalized, the matrix operation is used to obtain the similarity result matrix of the corresponding attribute.

Further, the format of the input data set in the processing module includes relational database form, NOSQL database form, ARFF file, CSV file, text file and excel spreadsheet.

Further, in the extraction module, the global information includes the number of records in the input data set and the number of data attributes in each record, and the local information includes the identifier, data type and value range of each data attribute.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (10)

1. A general similarity calculation method based on a heterogeneous information network is characterized by comprising the following steps:

step 1, preprocessing an input data set to ensure the validity of the input data;

step 2, extracting metadata, extracting description information of the input data, and storing the description information in a metadata database, wherein the description information comprises global information of the overall situation of the input data set, local information of each record, and conversion and corresponding information of an identifier and an internal representation of a data attribute;

step 3, selecting entities and data attributes participating in similarity calculation by a user, inquiring corresponding metadata, displaying the data type and value range of each metadata, prompting the user to select a processing method of the metadata according to a preset processing library, linking the entities according to semantic relations by taking the entities as central nodes, connecting the attributes to the corresponding entities, generating a heterogeneous information network mode, and storing the heterogeneous information network mode;

step 4, in the heterogeneous information network mode, after a user specifies data attributes and a meta path, similarity calculation is carried out according to the selected processing method;

and 5, fusing the plurality of similarity results obtained by calculation in the step 4 according to the weight of the selected data attribute to obtain a uniform similarity result, and performing format conversion on the uniform similarity result to output the uniform similarity result to a user.

2. The method for computing the similarity based on the universal heterogeneous information network according to claim 1, wherein in the step 1, the preprocessing of the input data set comprises data cleaning and data integration;

wherein,

the data cleaning is to perform format conversion, eliminate noise and delete inconsistent data, perform data cleaning on the input data set, remove useless noise data and perform corresponding format conversion;

data integration is the combining of multiple data source data.

3. The method for calculating the similarity based on the universal heterogeneous information network according to claim 1, wherein the step 4 specifically comprises:

step 401, a user designates data attributes and selects a meta path, and if a plurality of data attributes are involved, the user designates a link sequence of the plurality of data attributes to form the meta path;

step 402, after a user designates a single data attribute, calling data in the corresponding data attribute, constructing the data into an adjacency matrix, if the user designates a meta-path, constructing the adjacency matrix by using all relations on the meta-path, and then performing standardization processing on the adjacency matrix;

step 403, after the normalization of the adjacent matrix, obtaining a similarity result matrix corresponding to the attribute by using matrix operation.

4. The method for computing the similarity of the universal heterogeneous information network according to claim 1, wherein the format of the input data set in step 1 comprises a relational database format, a NOSQL database format, an ARFF file, a CSV file, and a text file and an excel spreadsheet.

5. The method according to claim 1, wherein in step 2, the global information includes the number of records of the input data set and the number of data attributes in each record, and the local information includes an identifier, a data type, and a value range of each data attribute.

6. A general similarity calculation system based on a heterogeneous information network, comprising:

the processing module is used for preprocessing the input data set and ensuring the validity of the input data;

the extraction module is used for extracting metadata, extracting description information of the input data and storing the description information in a metadata database, wherein the description information comprises global information of the whole condition of the input data set, local information of each record, and conversion and corresponding information of an identifier and internal representation of a data attribute;

the modeling module is used for selecting entities and data attributes participating in similarity calculation by a user, inquiring corresponding metadata, displaying the data type and value range of each metadata, prompting the user to select a processing method of the metadata according to a preset processing library, taking the entities as a central node, linking the entities according to semantic relations, connecting the attributes to the corresponding entities, generating a heterogeneous information network mode and storing the heterogeneous information network mode;

the computing module is used for carrying out similarity computation according to the selected processing method after a user specifies data attributes and a meta path in the heterogeneous information network mode;

and the post-processing module is used for fusing the plurality of similarity results obtained by calculation in the step 4 according to the weight of the selected data attribute to obtain a uniform similarity result, and performing format conversion on the uniform similarity result to output the uniform similarity result to a user.

7. The universal heterogeneous information network based similarity calculation system according to claim 6, wherein the preprocessing of the input data set by the processing module comprises data cleaning and data integration;

wherein,

the data cleaning is to perform format conversion, eliminate noise and delete inconsistent data, perform data cleaning on the input data set, remove useless noise data and perform corresponding format conversion;

data integration is the combining of multiple data source data.

8. The universal heterogeneous information network based similarity calculation system according to claim 6, wherein the calculation module comprises:

the specifying module is used for specifying the data attributes by a user and selecting the meta path, and if the plurality of data attributes are related, the user specifies the link sequence of the plurality of data attributes to form the meta path;

the construction module is used for calling data in corresponding data attributes after a user designates a single data attribute, constructing the data into an adjacency matrix, constructing all relations on a meta path into the adjacency matrix if the user designates the meta path, and then standardizing the adjacency matrix;

and the operation module obtains a similarity result matrix corresponding to the attributes by utilizing matrix operation after the adjacent matrix is normalized.

9. The generalized heterogeneous information network based similarity calculation system according to claim 6, wherein the format of the input data set in the processing module comprises a relational database form, a NOSQL database form, an ARFF file, a CSV file, and a text file and an excel spreadsheet.

10. The system according to claim 6, wherein in the extraction module, the global information includes the number of records of the input data set and the number of data attributes in each record, and the local information includes an identifier, a data type, and a value range of each data attribute.