JP2020144846A - Method of searching web page and computer-readable storage medium - Google Patents

Abstract

To provide a method of searching a web page executed by at least one processor.SOLUTION: The method includes the steps of: receiving a piece of information in which the content of search waiting is limited; recognizing the entity involved in the information to determine the type and attribute of the entity; generating search conditions based on the entity type and attributes; and, searching a data set using the search conditions to acquire a web page address that meets the search conditions. The data set is formed by extracting a web page part that has a predetermined structure from each web page in the web page set, and the search is performed over the web page set.SELECTED DRAWING: Figure 9

Description

The present invention relates to information extraction and information retrieval technology, and more particularly to extraction of web page (Web) content and search for web page content.

With the rapid development of the Internet, fixed-access devices, and mobile access terminals, web pages have become the main medium for acquiring and creating information. However, with the rapid increase in the number of web pages, it is difficult to obtain necessary information quickly and accurately.

Web page information extraction is information extraction using a web page as an information source, that is, information extraction from a semi-structured Web document. The core is to extract hidden information points in semi-structured HTML pages that are distributed on the Internet and express them in a clearer sense and more structured format. This is very convenient for users to query data in Web documents and for application programs to directly use data in Web documents.

For now, the Search Engine (SE) is an important tool used for knowledge search. When the user enters a single query, the SE can find the set of web pages that match the query from the index, sort them, and return the web pages to the user. In addition, the user can further search for the desired content from the returned result.

Traditional SEs can meet most of the needs of users, but as network resources become more abundant, so do the needs of users. In some cases, traditional SEs cannot meet this increased need. For example, when the content a user wants is spread across multiple web pages, the SE needs to find a set of web pages to satisfy the user's search intent. For example, users can use conventional SE to find out what kind of mobile phone screen is larger than 5 inches, what kind of foreign-affiliated company is in Chiyoda Ward, and how many experts are in the field of device translation. When trying to get answers to questions such as "is it?", It is rare that only one web page obtained by a traditional SE will contain all the answers to these questions.

In order to solve the above-mentioned problems, the present invention provides a web page search method and a computer-readable storage medium that are improved from conventional search engines, and integrates information distributed on a network to make relative information. It is possible to provide a complete and accurate integration result.

According to one aspect of the invention, a web page search method performed by at least one processor is provided, which receives information with limited search awaiting content; recognizes the entity involved in that information, and of the entity. Determine the type and attributes; generate search conditions based on the type and attributes of the entity; and search the data set using the search conditions to obtain a web page address that satisfies the search conditions. Including, the data set is formed by web page portions having a predetermined structure extracted from each web page in the web page set, and the search is performed on the web page set.

According to another aspect of the present invention, a computer program storage medium for storing a program is provided. The program causes the computer to perform the web page search method as described above when executed by the computer.

It is a comparison diagram of the search result of the conventional SE and the integrated SE. It is a figure which shows an example of the search result which performs the exemplary search using the conventional SE. It is a figure which shows an example of the table information in an exemplary web page. It is a comparison figure of the method of performing an integrated search using the integrated SE in the Example of this invention, and the method of performing a search using the conventional SE. It is a figure which shows an example of the HTML style of a table in a web page. It is a figure which shows the DOM tree structure of the table information in the example web page shown in FIG. It is a figure which shows the conversion operation to the table in a web page. It is a figure which shows the exemplary search process in the Example of this invention. It is a flowchart of the web page search method in the Example of this invention. It is a block diagram of the exemplary system which can realize the Example of this invention.

Hereinafter, preferred embodiments for carrying out the present invention will be described in detail with reference to the attached drawings. It should be noted that such an embodiment is merely an example and does not limit the present invention.

FIG. 1 is a comparison diagram of search results between the conventional SE and the integrated SE.

SE has already matured for many years and has some intelligence. However, the search scope of SE is still narrow, that is, each web page obtained by a conventional SE-based search contains the content that the user wants. For example, when searching with "Fujitsu" as a keyword, it is possible to obtain a plurality of web pages such as the homepage of Fujitsu Limited and the Wikipedia page. Each web page contains a description of "Fujitsu", and the user can obtain the desired information by further selecting a certain web page. FIG. 1 is a diagram showing a comparison of search results between the conventional SE and the integrated SE. The circles in Figure 1 represent the user's search input, and as you can see by comparison, each web page obtained by a conventional SE-based search contains the user's search input. However, each web page obtained by searching based on the integrated SE (where the term integrated SE is named to distinguish it from the traditional SE) contains only a portion of the search results. included. While traditional SEs can meet most of the needs of users, as network resources become more abundant, so do the needs of users, and in some cases traditional SEs meet these increased needs. I can't respond. For example, when the content that a user wants is distributed over multiple web pages, the SE cannot meet the user's search intent without finding a set of web pages (b in Figure 1). As shown by). For example, users can use conventional SE to "what kind of mobile phone screen is larger than 5 inches", "what kind of foreign-affiliated company is in Chiyoda Ward", "how many experts in the field of device translation". When trying to get answers to questions such as "is it?", It is rare that only one web page obtained by a traditional SE will contain all the answers to these questions (Figure 2 is traditional). The search results (Chinese) by SE are shown, and FIG. 2 is a diagram showing an example (Chinese) of the search results obtained by performing an exemplary search using the conventional SE). In other words, similar information exists on different web pages, for example, different web pages contain mobile phones of different brands and models, as well as screen size information. A collection of these web pages constitutes the search results for a question, and such searches are referred to herein as integrated searches. That is, the answer by the conventional SE is page1 or page2 or… or pageN, and the answer by the integrated SE is page1 ＋ page2 ＋… ＋ pageN.

For now, users can use many websites, such as e-commerce platforms, to perform vertical searches (eg, product, gourmet information, etc.). However, for more other types of entities, such as mechanisms, people, chemicals, drugs, etc., there are still few platforms that can directly provide search services. Therefore, if we can integrate the information distributed on the network and provide one relatively complete integration result, we can not only bring about a better user experience, but also have a wider application foreground.

In order to solve the above-mentioned problems, in the present invention, by effectively recognizing and extracting the attributes of various entities from a web page, ordinary information retrieval can be performed, and the attributes of the entities can also be searched. A method is provided in which the web page set in which all the entities that match the search conditions are located can be fed back to the user as a whole.

FIG. 3 is a diagram showing an example (Chinese) of table information in an exemplary web page.

Many web pages on the Internet contain descriptions of objects such as people, mechanisms, and goods, and these objects are referred to as "entities." There is a large amount of physical information on the Internet in the form of tables (as shown in Figure 3). Compared to text, it is easier to work with tables, and tables often contain information about attributes about an entity or other attributes about an entity. Therefore, the reliability of the result obtained by performing the search based on the table is high. Hereinafter, in each embodiment, a table will be described as an example, but the present invention is not limited thereto. Web pages also have other structures that represent the attributes of an entity or related attributes, but a comprehensive list is omitted here. It should be understood that methods for extracting other structures can be realized by referring to the following exemplary methods for extracting tables.

FIG. 4 is a comparison diagram between the method of performing an integrated search using the integrated SE in the embodiment of the present invention and the overall processing flow of performing the search using the conventional SE.

When processing web page information, it is often necessary to extract tables from web pages on the web to recognize the attributes of the entity. However, traditional methods only process standard table elements in web pages, namely "table" elements. The "table" element is the standard table element for web pages, but many tables are made up of non- "table" elements. Processing only the "table" element can result in a lot of information being discarded. For example, Figure 5 shows the HTML format of a table on a web page. A in FIG. 5 is an HTML format of an example table which is a table in appearance as shown in FIG. 3, and the table is composed of elements of “dl”, “dt”, and “dd”. B in 5 is a table of normal "table" elements. Since it is possible to construct an apparent table with arbitrary HTML elements, it is not possible to create a template in advance. Therefore, the usual method for standard table elements in a web page cannot handle such an unusual table, which is a table in appearance. However, the inventor has found that tables that describe the attributes of an entity often have two important properties:

1) There is an attribute name in the table. For example, as shown in Figure 3, the table has attribute keywords such as "price" (5388 yuan, 6388 yuan) and "color" (gold, silver, rose gold, black, bright black, red). In addition, since the attributes of a given entity can be listed within a finite range, it is possible to find the position where the attribute appears by forming an attribute keyword and then performing matching in the HTML page. .. In addition, the place where the attribute name is given is likely to be the position of the table.

2) The HTML element of the table has a repeating structure. As shown in FIG. 5, the element corresponding to each row has a repeating feature, regardless of whether the table is composed of "table" elements. For example, in a in FIG. 5, each row in the table corresponds to two elements "dd, dt", and the "dd, dt" elements repeatedly appear under the "dl" node, and also in the figure. In b in 5, the "tr" element in each row repeats in the same way and appears under the "tbody" node.

Therefore, it is conceivable to find a possible position of the table by matching the names of the attributes, and to determine the position of the table by special conquest of the repeating structure. Therefore, first, it is necessary to construct one dictionary that describes the attributes of the entity.

<Construction of attribute dictionary>
For now, there are already some retrievable databases that provide a description of the attributes. Therefore, it is possible to construct a dictionary of attribute keywords from a plurality of conventional databases. For example, the example in FIG. 3 is derived from "Baidu Baike", and the example in b in FIG. 5 is derived from "Wikipedia". These conventional resources are easy to obtain, of good quality, and have classification information.

First, one dictionary D = {cat ₁ , cat ₂ , ..., cat _n } is formed, of which cat _i represents a type, that is, a type of entity, such as a person, mechanism, or drug. Each type is represented as cat = [<term ₁ , value ₁ >, <term ₂ , value ₂ >,…, <term _n , value _n >], of which term represents the attribute keyword and value is Represents a list of attribute values. For example, the following dictionaries can be obtained based on the examples in FIGS. 3 and 5.

D = {['Handset (Chinese. Meaning is celluar phone. Same below)': [<'Promulgation (Chinese. Meaning is release. Same below) Time','…'>, <'Price', '…'>, <'Complex (Chinese. Meaning is color. Same below)','…'>, ……, <'First release (Chinese. Meaning is first release. Same below)','…'>,<'Type','…'>,<'Scale (Chinese. Meaning is size. Same below)','…'>, ……]]}
Early-form dictionaries are localized. That's because it may not include all the attribute keywords on a web page. Therefore, it is necessary to extend the dictionary. For each type of cat, perform a CWD (Chinese Word Distinguishing) operation on the attribute keyword term in it, and then stat the frequency of words with a length greater than 1, and the frequency is predetermined. Insert a famous word or a moving word that is larger than the number of times (predetermined threshold) into the dictionary as the root of the attribute. In the case of the above example, if a word with a frequency higher than 1 is set as the root of the attribute, the root "promulgation" (frequency = 2) can be obtained, and it is described as "% promulgation%". An attribute value is a set of values corresponding to all attributes including the root. When it is inserted into the dictionary, the dictionary looks like this:

D = {['Handset': [<'Promulgation time','…'>, <'Price','…'>, <'Complex','...'>, ……, <'Primary promulgation','…'>,<'Type','…'>,<'Scale','…'>,<'%Promulgation%','…'> ……]]}
The action of the root is to show that some attributes are related to the root. In addition, multiple expression methods, for example, "promulgation time" and "primary promulgation" may be related to "promulgation", and if there are other expression methods including "promulgation", the same concept may be expressed. There is also. Therefore, fuzzy matching of attribute names can be performed using roots.

The dictionary can be constantly updated in the manner described above. In addition, the generated dictionary is used for each subsequent operation.

<Extracting tables in web pages>
The following describes how to extract the table in the web page. To get one HTML web page, first convert the HTML page to a DOM tree. FIG. 6 is a diagram showing a DOM tree structure of table information in an exemplary web page in FIG. For convenience, some table nodes and all character nodes are omitted in Figure 6. The nodes at the bottom of the DOM tree are considered leaf nodes. For each node in the DOM tree, generate a path pattern and write it as pat. The path pat of each node is composed of the node names of all the nodes passing from the node to the Nth ancestor node and the class attribute of the node. For example, when N = 2, the pat of “dl” in FIG. 6 is as follows.

pat = dl-basicInfo-block-basicInfo-left_div-basic-info-cmn-clearfix_div
For each leaf node, the characters at the leaf node are matched in the dictionary D generated as described above, and if the same character is obtained by matching, the node is recognized as an anchor node. To do. In FIG. 6, the dark leaf node is the anchor node obtained by matching.

After determining the anchor node, for each non-leaf node, select a table node to locate the table based on the following two criteria.

1) The child node has a pat that appears repeatedly, and the number of times it appears repeatedly is t1 or more;
2) Of the leaf nodes it covers, there are at least t2 anchor nodes (covered).

The values of t1 and t2 may be set according to the needs, and may be set based on, for example, the type of the target waiting for search, or may be set regardless of the type of the target. When the value of t1 is 3 and the value of t2 is 3, "dl" is selected as the table node, and among its child nodes, pat of dt + dd appears three times repeatedly, and also. , To cover 3 anchor nodes.

After determining the table node, that is, after determining the position of the table, it is necessary to subsequently determine the format of the table. Based on the format of the table on the web page, the table can be divided into the following two types.

1) Vertical structure: The first row is the table header, and the second to last rows are the corresponding values;
2) Left and right structure: The first column is the table header, and the second to last columns are the corresponding values.

The reason for this division is that the tables are usually aesthetically horizontal or vertical tables. Therefore, by dividing as described above, all tables having these two types of table formats can be recognized.

Determining the format of a table is, for example, determining the number of rows and columns of a table, and the calculation of the number of rows and columns of a table requires converting the table to an internal display format. .. The number of rows may be the number of repetitions of the path pat, and as shown in FIG. 6, since “dt + dd” appears three times repeatedly, the table has three rows and the number of columns. For, the number of child nodes of the node corresponding to the route pat may be calculated, and then the maximum number of child nodes in all rows may be set as the number of columns. Taking FIG. 6 as an example, each pat corresponds to two nodes "dd" and "dt", and each of these two nodes has one child node. Therefore, the number of columns is 2 because the number of child nodes corresponding to each row is all 2. This gets one 3 * 2 table and then inserts the HTML node into the corresponding unit. The process is shown in FIG.

<Table classification and attribute extraction>
After performing the above conversion operation on the table, the table is classified, that is, the type of entity described in the table is obtained. For example, it is necessary to know that the entity described in the table in Fig. 3 is a "hand desk". First, from the table after the conversion operation, the special conquest vector feat = [[term ₁ , [v ₁₁ , v ₁₂ , ..., v _1n ]], [term ₂ , [v ₂₁ , v ₂₂ , ..., v _2n ]] ],…, [term _m , [v _m1 , v _m2 ,…, v _mn ]]], of which term is the attribute element in the table header and v represents the value corresponding to the attribute. .. Taking the table in Fig. 7 as an example, feat = [[dt ₁ , [dd ₁₁ ]], [dt ₂ , [dd ₂₁ ]], [dt ₃ , [dd ₃₁ ]]] can be obtained. Here, for convenience, it is expressed using dt and dd, but it is actually the text at the corresponding leaf node. For each type cat in Dictionary D, calculate the cosine similarity with feat by the following formula (1).

Of these, | a | represents the number of vocabularies in a, and sim (a, b) can be calculated by the following formula (2).

Of these, t _cat represents a list of all attribute element terms contained in the type cat, t _feat represents the table header attribute list in the table special conquest vector, and v _cat represents all corresponding attribute element terms in the type cat. Indicates the value list of, v _feat represents the list of all values in the special conquest vector, α and β each represent the weight of these two parts, and α + β = 1, tsim. (*) Can be calculated by the following formulas (3) and (4).

Of these, l is a list of words, w1 and w2 are words in the two lists waiting to be calculated, respectively, and ssim (*) calculates the similarity between the two strings. When the two strings are exactly the same, the similarity is 1, and when some are the same (fuzzy matching), it is 0.6.

After calculating the similarity, the type cat having the maximum value (and larger than a predetermined threshold value) may be used as the table type. With table types, triples (sets of three elements) can be generated. The generated triples are stored in a remote or local database and are used, for example, to be queried by sparql. For example, by performing the above processing on the table in FIG. 7, a triple having the following format can be obtained.

page_url category cat
page_url dt1 dd1
page_url dt2 dd3
page_url dt3 dd3
Among them, page_url indicates the url of the current web page, category is the classification result of the table, and dt and dd are all characters in the web page. By using the triple generated as described above, it is possible to query and search and query the corresponding url set.

FIG. 8 is a diagram illustrating an exemplary search process in an embodiment of the present invention.

First, the user sets a query phrase template for search in advance. As shown in Figure 8, the horizontal line in the template is called the position slot, and the process of generating the search term actually extracts information from the user-entered query query and fills the position slot. It's a process. First, using the dictionary D generated as described above, match the type cat related to the query query and obtain it; after obtaining the type cat by the query, fill the position slot of the classification result category of the table. After that, match the attributes in the query query using the attribute element term list corresponding to the type cat, and if multiple are obtained by matching, generate multiple search conditions and "or" between the conditions. Finally, the attribute value in the condition is replenished, and then the search is performed, and the result obtained by the search is an HTML url set, which is further converted into an HTML page and returned to the user ( You can also return). As shown in FIG. 8, when filling the attributes, the attribute name and the attribute value generated based on the meaning may be used. For example, "screen" may be "screen size". “Greater than 5 inches” may be converted to “> 5 inches”. By such conversion, more accurate and complete search results can be obtained at the time of search. Figure 8 shows a select template that translates a user's query query into a sparql language, and of course, the user's search content input is not limited to sparql queries and may be translated into queries based on other languages. ..

FIG. 9 is a flowchart of the web page search method according to the embodiment of the present invention.

The method starts at step 901. In step 901, information with a limited search wait content is received, for example, the user-entered "mobile phone with a screen larger than 5 inches" shown in FIG. 8 is received; subsequently, in step 902, the information is input. It recognizes the entity involved in the information, and the type of entity (eg, "mobile phone" in FIG. 8) and attributes (attributes include attribute names and attribute values, eg, "screen", "" in FIG. "Screen size", "> 5 Lynch") are confirmed, and specifically, the type and attribute of the entity related to the search waiting content are used using the template (Dictionary D) of the relationship between the type of entity and the attribute described above. In step 903, search conditions are generated based on the type and attributes of the entity, for example, as shown in FIG. 8, three parallel search conditions are generated; in step 904, the search conditions are used. Then, a search is performed on the data set to obtain a web page address that satisfies the search conditions, and the data set has a predetermined structure extracted from each web page in the web page set in which the search is performed. The data set formed by the web page portion having may be, for example, a set consisting of each triple extracted as described above, and specifically, the relationship between the type of entity and the attribute as described above. A template (Dictionary D) is used to extract a web page portion having a predetermined structure and further convert the web page portion into a triple, in which the triple is the type of entity, the attribute of the entity and the entity. Used to represent the relationships between web page addresses that correspond to a type or attribute of, the data set consists of triples that correspond to different entity types.

According to the embodiment of the present invention, a template (dictionary described above) of the relationship between the type of entity and the attribute is obtained from an information source having classification information.

According to the embodiment of the present invention, the predetermined structure is a table structure adjusted in a predetermined direction, and the predetermined direction is a horizontal direction or a vertical direction.

According to an embodiment of the present invention, a remote server can generate a set of triples, in which case providing a query interface for a local user to query the remote server.

According to each embodiment of the present invention, a set of triples can be generated on a local computer, machine, host, etc., whereby a user can execute a query on the local computer, machine, host, etc.

Further, the above-mentioned series of processes may be realized by software and / or firmware. When implemented by software and / or firmware, install the programs that make up the software from a storage medium or network on a computer with a dedicated hardware structure, such as the general-purpose machine 1000 (eg, computer) shown in FIG. , The computer can perform various functions and the like when various programs are installed.

FIG. 10 is a block diagram of an exemplary system that can realize the embodiments of the present invention.

In FIG. 10, the central processing unit (CPU) 1001 performs various processes based on the program stored in the ROM 1002 or the program rodged from the storage unit 1008 to the RAM 1003. The RAM 1003 can also store data required when the CPU 1001 performs various processes according to needs. CPU 1001, ROM 1002 and RAM 1003 are connected to each other via buzz 1004. The input / output interface 1005 is also connected to the bus 1004.

Further, the following components are connected to the input / output interface 1005, that is, an input unit 1006 including a keyboard and the like, an output unit including a display such as a liquid crystal display (LCD), and a speaker. 1007, storage unit 1008 including hard disk, etc., communication unit 1009 including network interface card, for example, LAN card, modem, etc. The communication unit 1009 performs communication processing via a network such as the Internet or LAN.

Drive 1010 may be connected to input / output interface 1005, if desired. The removable medium 1011 such as a semiconductor memory can be set in the drive 1010 as needed, and the computer program read from the medium can be installed in the storage unit 1008.

The present invention further relates to a computer storage medium in which the program is stored, the program being able to realize the method of the above-described embodiment when executed.

The present invention also provides a program product that includes a machine-readable command code. When such a command code is read and executed by the machine, the method according to the embodiment of the present invention described above can be executed. Correspondingly, carry such program products, such as magnetic disks (including floppy disks (registered trademarks)), optical disks (including CD-ROMs and DVDs), magneto-optical disks (MD (registered trademarks)). ), And various storage media such as semiconductor storage devices are also included in the present invention.

The above-mentioned storage medium may include, but is not limited to, for example, a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor storage device, and the like.

Further, each operation (process) in the above method can be realized by a method of a computer-executable program stored in various machine-readable storage media.

In addition, the above examples and the like will be further disclosed as additional notes as follows.

(Appendix 1)
A web page search method performed by at least one processor,
Received information with limited search waiting content;
Recognize the entity related to the information and determine the type and attributes of the entity;
A search condition is generated based on the type and attribute of the entity; and a search is performed on a data set using the search condition to obtain a web page address satisfying the search condition.
A method in which a web page portion having a predetermined structure is extracted from each web page in a web page set to form the data set, and the search is performed on the web page set.

(Appendix 2)
The method described in Appendix 1
A method for determining the type and attribute of an entity related to the search waiting content by using a template of the relationship between the type of entity and the attribute.

(Appendix 3)
The method described in Appendix 1
A method of extracting a web page portion having the predetermined structure by using a template of a relationship between an entity type and an attribute.

(Appendix 4)
The method described in any one of Supplementary notes 1 to 3.
Extracting a web page portion having a predetermined structure from each web page in the web page set to form the data set can be performed.
Including converting the web page portion to a triple
The triple is used to represent the type of entity, the attributes of the entity, and the relationship between the type of entity or the web page address corresponding to the attribute.
A method in which the data set consists of triples corresponding to different entity types.

(Appendix 5)
The method described in Appendix 2 or 3,
A method of obtaining a template of a relationship between the type of an entity and an attribute from an information source having classification information.

(Appendix 6)
The method described in any one of Supplementary notes 1 to 3.
A method in which the predetermined structure is a table structure.

(Appendix 7)
The method described in Appendix 6
A method in which the predetermined structure is a table structure after adjustment in a predetermined direction, and the predetermined direction is a horizontal direction or a vertical direction.

(Appendix 8)
The method described in any one of Supplementary notes 1 to 3.
A method in which the attribute comprises an attribute name and an attribute value.

(Appendix 9)
The method described in any one of Supplementary notes 1 to 3.
Generating search conditions based on the type and attributes of the entity
Multiple parallel search conditions are generated using expressions similar to the meanings of the substance types and attributes, and when searching, the multiple parallel search conditions are connected and searched using'or'. A method that involves forming a phrase.

(Appendix 10)
The method described in any one of Supplementary notes 1 to 3.
Generating search conditions based on the type and attributes of the entity
A method comprising converting the entity type and attributes into a query phrase template.

(Appendix 11)
The method described in Appendix 10
A method in which the query phrase template is a sparql query phrase template.

(Appendix 12)
The method described in any one of Supplementary notes 1 to 3.
A method in which a template for the relationship between the types and attributes of the entity is generated based on "Baidu Baike" and / or "Wikipedia".

(Appendix 13)
A computer-readable storage medium that stores programs
The program is a storage medium that, when executed by a computer, causes the computer to perform the method according to any one of Appendix 1-12.

Although the preferred embodiment of the present invention has been described above, the present invention is not limited to this embodiment, and any modification to the present invention belongs to the technical scope of the present invention unless the gist of the present invention is abandoned.

Claims (10)

A web page search method performed by at least one processor,
Received information with limited search waiting content;
Recognize the entity related to the information and determine the type and attributes of the entity;
A search condition is generated based on the type and attribute of the entity; and a search is performed on a data set using the search condition to obtain a web page address satisfying the search condition.
A method in which a web page portion having a predetermined structure is extracted from each web page in a web page set to form the data set, and the search is performed on the web page set. The method according to claim 1.
A method for determining the type and attribute of an entity related to the search waiting content by using a template of the relationship between the type of entity and the attribute. The method according to claim 1.
A method of extracting a web page portion having the predetermined structure by using a template of a relationship between an entity type and an attribute. The method according to any one of claims 1 to 3.
Extracting a web page portion having a predetermined structure from each web page in the web page set to form the data set can be performed.
Including converting the web page portion to a triple
The triple is used to represent the type of entity, the attributes of the entity, and the relationships between the types or attributes of the entity and the web page addresses that correspond to them.
A method in which the data set consists of triples corresponding to different entity types. The method according to claim 2 or 3,
A method of obtaining a template of a relationship between the type of an entity and an attribute from an information source having classification information. The method according to any one of claims 1 to 3.
A method in which the predetermined structure is a table structure. The method according to claim 6.
A method in which the predetermined structure is a table structure after adjustment in a predetermined direction, and the predetermined direction is a horizontal direction or a vertical direction. The method according to any one of claims 1 to 3.
A method in which the attribute comprises an attribute name and an attribute value. The method according to any one of claims 1 to 3.
Generating search conditions based on the type and attributes of the entity
A plurality of parallel search conditions are generated using expressions similar to the meanings of the substance types and attributes, and at the time of search, the multiple parallel search conditions are connected and searched using "or". A method that involves forming a phrase. A computer-readable storage medium that stores a program for causing a computer to execute the method according to any one of claims 1 to 9.

Images