CN101916272B - A Data Source Selection Method for Deep Web Data Integration - Google Patents

Abstract

The invention discloses a data source selection method for deep web data integration, which comprises the following steps of: firstly, selecting a deep web data source which has relatively high correlation degree with customer query based on query interface semantic features and by combining an ontology library; secondly, evaluating the quality of data sources through the quality evaluation model of the data sources; and finally, acquiring a data source set which has high correlation degree with the customer query and higher quality according to the quality evaluation condition. Compared with the prior art, the method can improve the accuracy of deep webpage query and reduce information redundancy and improve query efficiency at the same time.

Description

A Data Source Selection Method for Deep Web Data Integration

technical field

The invention relates to a network-based data source selection method, in particular to a deep network data source selection method connected by a network query interface, which is used for integration services of deep network data sources.

Background technique

With the wide application of network database, the network is accelerating the "deepening". There are a large number of pages on the Internet that are dynamically generated by the background database. This part of the information cannot be obtained directly through static links, but can only be obtained by filling out forms and submitting queries. Since traditional web crawlers (Crawlers) do not have the ability to fill out forms, they cannot crawl. these pages. Therefore, existing search engines cannot search for this part of the page information, resulting in this part of the information being hidden and invisible to users, which is called the Deep Web (Deep Web, also known as Invisible Web, Hidden Web). Deep Web is a concept corresponding to Surface Web, originally proposed by Dr. Jill Ellsworth in 1994, referring to those Web pages whose information content is difficult to find by common search engines. Deep Web information is generally stored in a database and needs to be accessed through a query interface. Compared with static pages, it usually has a larger amount of information, more specific topics, better information quality, better information structure, and faster growth. Research shows that there is 500 times more information on the Deep Web than on the Surface Web, with nearly 450,000 Deep Web sites. Realizing large-scale Deep Web data integration is an effective way to facilitate users to use Deep Web information.

The large-scale Deep Web integration system mainly includes: 1) Deep Web Discovery; 2) Query Interface Extraction; 3) Source selection; 4) Query Transfer; 5) The five key parts of Result Merging.

Deep Web data sources include data resources on a variety of topics, and there are many Deep Web data sources on a certain topic. These data sources belong to the same topic, but the data quality varies greatly: some are outdated, inaccurate, or inconsistent. , while some are updated in a timely, accurate and consistent manner. Moreover, these data sources contain different amounts of data, covering each other, some covering a large amount, and some even completely containing other data sources. Taking business and education as an example, according to the statistics of Complete Planet, there are thousands of Web databases. Since Complete Planet only collected about 7% of the Web databases in the entire Deep Web data source, it is far behind in reality. Much larger than this number (Bergman M.K. The Deep Web: Surfacing Hidden Value. In Journal of Electronic Publishing, 2002, 7(1): 8912-8914). Kabra G et al proposed a method of selecting (Top-k) k Deep Web data sources closest to the content of the user's query request for query (KabraG, Li CK, Chang KCC. Query routing: Finding Ways in the Maze of the Deep Web. In Proc. of the ICDE, 2005, 64-73). The above methods only deal with the simple attribute relationship of the query interface, and query the form through keywords. These methods do not take into account the semantic relationship between the attributes of the query interface, and the accuracy of the data source selection result in the process of selecting the corresponding data source is low, and The returned data source result is not equal. With the continuous growth of the number of Web databases, the selection of Deep Web data sources has become a key issue that needs to be solved urgently.

Contents of the invention

The purpose of the present invention is to provide an efficient and accurate method for selecting deep network data sources to improve the selection efficiency and accuracy of deep network data sources.

Data source selection refers to selecting a query interface set whose relevance to user query is greater than a set threshold or selecting a query interface set with a larger correlation value under the condition of a given Deep Web data source query interface set and a certain user query. The process of querying the interface set of k data sources. The main purpose of data source selection is to select databases with high coverage and small overlap to avoid a large amount of redundant and irrelevant information; users hope to find corresponding high-quality query results, and also hope to obtain comparisons between the same results. Most existing data source selection methods directly calculate the correlation between user queries and query interfaces to perform keyword matching. Due to the following three reasons, when using existing methods, user queries are usually inaccurate and have high redundancy. redundancy while discovering some irrelevant data sources:

Firstly, because there are a large number of accessible Deep Web resources in the same field, accessing a large number of Deep Web resources on the Internet is a time-consuming and laborious process; secondly, the data quality of each database varies greatly, and some are outdated, inaccurate or inconsistent. And some are timely, accurate and consistent. Not every Deep Web can satisfy a specific query. Obviously, the Deep Web in any field cannot contain all the information in this field, so it is impossible to satisfy any query in this field. In the end, most of the Deep Web data sources in a field contain data of different sizes, covering each other, some covering a large area, and even completely including other data sources; and there is redundant information between them, For a query, the more visits to the Deep Web, the greater the redundancy of the returned information, which greatly increases the difficulty of processing redundant information.

Based on the above analysis, it can be seen that the goal to be achieved in the selection of Deep Web data sources is how to select a suitable subset from a large number of Deep Web data sources in a field, reduce the number of access to Deep Web and make the query results redundant. The margin is small enough, and the query cost is lower.

To this end, we use the semantic features of the query interface to extend the user query based on the domain ontology, so that the selected query interface set can better meet the user's query requirements. Specifically, the technical scheme of the present invention is as follows:

A data source selection method for deep web data integration, characterized in that it comprises the following steps:

Step A, analyzing the query interface;

Step B, construct ontology library and transform corresponding query information into ontology information through ontology library;

Step C, calculate the correlation degree between the ontology information and each data source, and select the data source that meets the preset conditions according to the correlation degree; for a given target query interface object DWI _i and query ontology Q _i , the correlation degree is calculated according to the following formula:

$\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; DWI</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; DWI</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}{\sqrt{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; DWI</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; \&times;</span>\sqrt{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; ,</span>$

Among them, R(DWI _i , Q _i ) represents the correlation between the query ontology Q _i and the query interface object DWI _i , and m is the number of objects in the query interface.

Ontology is a complex model with more semantic and structural information. The ontology library in the above step B can use the existing public ontology library; it can also collect the existing public ontology library and expand these ontology libraries. A new ontology library is obtained; and the present invention adopts the latter.

The main task of this type of ontology learning is to analyze the semantic information contained in the relational model and map it to the corresponding parts in the ontology. Secondly, the query interface and the data source result page usually contain rich information such as concepts, examples, and the relationship between domain-related concepts. The query interface appears in the form of an HTML form. If the database schema cannot be obtained, the HTML form can be analyzed. structure and data to obtain the semantics in the Web database, so as to build ontology. According to the above analysis, the ontology library of the present invention can be constructed through the following steps:

Step B1, analyze the HTML form schema structure through the existing ontology library to obtain the semantics of the query interface, and construct the classes in the corresponding ontology library;

Step B2, extract concepts and examples from the query interface and the result page, and extract the hierarchical relationship and functional relationship of classes in the existing ontology library;

Step B3. Extract the relationship between the ontology classes obtained in the above step B2 from multiple data sources of a certain subject, then reason and map different relationships, and finally merge into a higher-level domain ontology; for each ontology library For each class, build a set of keywords corresponding to the class to form the vocabulary layer of the ontology library.

In order to further improve the accuracy of data source selection, reduce information redundancy, and reduce query costs; the present invention introduces the concept of data source quality score on the basis of the above technical solution, and measures the quality of data source by the quality score of data source , select some data sources with higher quality scores and discard other data sources with lower quality, thus greatly reducing information redundancy and improving query accuracy. Specifically, after the above step C, continue to perform the following steps:

Step D, establishing a data source quality assessment model and using the data source quality assessment model to calculate the quality score of each data source obtained in step C;

Step E, select several high-quality data sources according to the quality score and according to a certain method, and obtain the final data source set.

The selection of several high-quality data sources based on the quality score and according to a certain method in the above step E may be to select a data source with a quality score greater than a preset threshold; the Top-k data selection method may also be used, that is, according to the quality score Sort the data sources from large to small, and select the top k data sources, where k is the preset number of finally selected data sources.

The method of the present invention first selects a deep network data source that is highly relevant to the user query based on the semantic features of the query interface and in combination with the ontology database; then measures the quality of the data source by the quality score of the data source, and selects some data with higher quality scores sources and discard other lower-quality data sources, and finally obtain high-quality data sources that are highly relevant to customer queries. Compared with the prior art, the method of the invention can improve the accuracy of deep web page query, reduce information redundancy and improve query efficiency.

Description of drawings

Fig. 1 is an example diagram of a deep web page query interface of a specific embodiment of the present invention;

Fig. 2 is a flow chart of the inventive method;

Fig. 3 is an example diagram of ontology library structure;

Detailed ways

The technical scheme of the present invention is described in detail below in conjunction with accompanying drawing:

As shown in accompanying drawing 2, the present invention carries out the selection of deep network data source according to the following steps:

Step A, analyzing the query interface;

As shown in Figure 1, a query interface includes some form controls for users to input query information, such as text box (Textbox), radio button (Radio Button), check box (Check box) and drop-down list (SelectionList) and other controls . Each control is usually associated with a label - a descriptive text, each control can have one or more values (value), for example a drop-down list has a list of values for the user to choose from, radio buttons and check boxes usually have a value . Logically speaking, a control and its associated label constitute an attribute, corresponding to a field in the deep web (Deep Web) background database. Typically, a property contains a label and one or more form controls. By parsing the current Deep Web query interface page, the labels and form controls corresponding to the content of each attribute are obtained, and then they are combined into attributes (a logical unit of query conditions) according to the semantic relationship. We can abstractly express the query interface ontology instance DWI as: DWI=(S, P, M). Among them, S reflects the specific information such as the function of the interface instance, which includes basic information such as the name of the interface instance (form tag name) and the URL of the interface site. P={p ₁ , p ₂ , . . . , p _n } is the ontology instance template corresponding to the interface instance, and M is the method provided by the interface instance. After the DWI instance is established, the user can provide an ontology instance-oriented query to retrieve the information he needs.

The Deep Web data source interface set can be abstracted as: Assume that the Deep Web data source interface set in a certain domain is DWS={S _i1 , S _i2 ,...,S _im }, each data source interface S _ii corresponds to a query interface The data source ontology template composed of instances R _i above, the union of all instances in the ontology template is the data source interface set DWS. The so-called instance refers to the tag name, internal attribute name, one or more modifiers and their value fields corresponding to an element on the specified query interface, which is the smallest semantic unit on the query interface.

Step B. Construct an ontology database and convert the corresponding query information into ontology information through the ontology database; wherein the construction of the ontology database is performed according to the following steps:

Step B1, analyze the HTML form schema structure through the existing ontology library to obtain the semantics of the query interface, and construct the classes in the corresponding ontology library;

Step B2, extract concepts and examples from the query interface and the result page, and extract the hierarchical relationship and functional relationship of classes in the existing ontology library;

Step B3. Extract the relationship between the ontology classes obtained in the above step B2 from multiple data sources of a certain subject, then reason and map different relationships, and finally merge into a higher-level domain ontology; for each ontology library For each class, build a set of keywords corresponding to the class to form the vocabulary layer of the ontology library;

The method of the present invention abstracts and expresses corresponding query information as a kind of query model: Deep Web represents the relational table DB that is made up of a series of query interface attributes: Aq={aq ₁ , aq ₂ ,..., aq _n } (interface mode) and a Attribute composition of series query results: Ar={ar ₁ , ar ₂ ,, ar _m } (result pattern). Among them, each attribute aq _i ∈ A represents the query attribute obtained through the query interface, and the result attribute arj ∈ A represents the attribute in the query result. Each query operation can be expressed by a similar SQL statement: "Select ar ₁ , ar ₂ ,, _{arm m} from DB WHERE aq1=val q ₁ , aq ₂ = valq ₂ ,..., aq _n = valq _n ", where val q _i represents the attribute value populated in the query form.

For the query information, a series of query interface sets are obtained through the query expansion of the ontology library. The ontology structure is shown in Figure 3, which is a part of the structure diagram of the ontology library with a vehicle (Vehicle) as the core concept. The ontology library structure includes a series of abstractions to real things. For example, concepts such as "Vehicle", "Car", and "Truck" constitute the classes in the ontology library. The figure also includes the relationship between classes such as "driver" and "price". The ontology library also includes Contains various corresponding entities, such as BWM, F512M, etc. Through the extension of the ontology library, a concept can be expanded into a series of concept sets in the ontology layer. For example, for the concept "Vehicle", its corresponding concepts also include concepts such as "Car" and "Truck".

Step C, calculate the correlation degree between the ontology information and each data source, and select the data source that meets the preset conditions according to the correlation degree; for a given target query interface object DWI _i and query ontology Q _i , the correlation degree is calculated according to the following formula:

$\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; DWI</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; DWI</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}{\sqrt{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; DWI</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; \&times;</span>\sqrt{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; ,</span>$

Among them, R(DWI _i , Q _i ) represents the correlation between the query ontology Q _i and the query interface object DWI _i , and m is the number of objects in the query interface.

Step D, establishing a data source quality assessment model and using the data source quality assessment model to calculate the quality score of each data source obtained in step C;

It can be known by analysis that the main factors affecting the evaluation of Deep Web data source quality are: browser, Web database, user and network performance. This specific implementation method uses these four types of factors as a first-level quality factor; each first-level quality factor includes Several second-level quality factors, for example, as a first-level quality factor, the Web database includes several second-level quality factors such as domain integrity, consistency, redundancy, and data source size, so that a two-level quality factor can be obtained The quality factor set, and the data source quality evaluation model is obtained as follows:

${\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; the\; s</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; K</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; \&times;</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; L</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; q</span>}}_{\mathrm{<span\; class="notranslate">\; nj</span>}}<span\; class="notranslate">\; \}</span>$

w_j为第n个一级质量因子中第j个二级质量因子的权重，q_nj为使用第n个一级质量因子中第j个二级质量因子评估第s个数据源的质量得分，j＝1，2…L，L为质量因子集中第n个一级质量因子中所包含二级质量因子的个数，

Among them, Q _s ∈ [0, 100], represents the quality score of the sth data source; W _n represents the weight of the nth first-level quality factor in the quality factor set, n=1, 2...K, K is the quality factor set The number of first-order quality factors,

w _j is the weight of the jth secondary quality factor in the nth primary quality factor, q _nj is the quality score of the sth data source evaluated using the jth secondary quality factor in the nth primary quality factor, j=1, 2...L, L is the number of secondary quality factors contained in the nth primary quality factor in the quality factor set,

The above data source quality assessment model is an existing technology, and more detailed content can be referred to in the literature (Xian Xuefeng, Fang Wei, etc. A Deep Web data source quality assessment model. Microelectronics and Computers, 2008, Vol 25(10): 47-50.).

Step E, select several high-quality data sources according to the quality score and according to a certain method, and obtain the final data source set.

In this specific embodiment, the Top-k data selection method is adopted in this step, that is, the data sources are sorted from large to small according to the quality score, and the first k data sources are selected, and k is the number of preset final selected data sources. number.

Claims (2)

1. A data source selection method for deep web data integration, characterized in that, comprising the following steps: Step A, analyzing the query interface; Step B, construct ontology library and convert corresponding query information into ontology information through ontology library; said building ontology library specifically follows the following steps: Step B 1. Analyze the HTML form schema structure through the existing ontology library to obtain the semantics of the query interface, and construct the classes in the corresponding ontology library; Step B2, extract concepts and examples from the query interface and the result page, and extract the hierarchical relationship and functional relationship of classes in the existing ontology library; Step B3. Extract the relationship between the ontology classes obtained in the above step B2 from multiple data sources of a certain subject, then reason and map different relationships, and finally merge into a higher-level domain ontology; for each ontology library For each class, build a set of keywords corresponding to the class to form the vocabulary layer of the ontology library; Step C, calculate the correlation degree between the ontology information and each data source, and select the data source that meets the preset conditions according to the correlation degree; for a given target query interface object DWI _i and query ontology Q _i , the correlation degree is calculated according to the following formula: $\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; DWI</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; DWI</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}{\sqrt{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; DWI</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; \&times;</span>\sqrt{\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{{\mathrm{<span\; class="notranslate">\; Q</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; ,</span>$ Among them, R(DWI _i , Q _i ) represents the correlation between the query ontology Q _i and the query interface object DWI _i , and m is the number of objects in the query interface; Step D, establishing a data source quality assessment model and using the data source quality assessment model to calculate the quality score of each data source obtained in step C; Step E, select several high-quality data sources according to the quality score and according to a certain method, and obtain the final data source set. 2. as claimed in claim 1, is used for the data source selection method of deep network data integration, it is characterized in that, according to quality score described in step E and select some high-quality data sources according to certain method means: according to quality score will The data sources are sorted from large to small, and the top k data sources are selected; k is the preset number of finally selected data sources.

Images