JP2011180824A - Device and method for imparting word meaning classified content, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device for imparting word meaning classified contents, imparting the appropriate contents such as images and moving images, to respective word meanings, when there is a direction word and a definition sentence indicating the word meaning or text similar to it, and widely covering the plurality of word meanings even for minor word meanings. <P>SOLUTION: For the direction word, a list wherein a plurality of retrieval word candidates to which relation with the direction word is attached respectively are sectioned for each word meaning is input, and a retrieval word selection part selects and outputs at least one retrieval word for each word meaning on the basis of the relation. A content retrieval part acquires predetermined contents for each word meaning of the direction word by retrieving a communication network using the retrieval word corresponding to the word meaning. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a natural language analysis technique or a content search technique such as an image or a video, and in particular, when there is a word and a definition sentence indicating the meaning or a text similar thereto, obtains appropriate content in accordance with the meaning of the word, The present invention relates to a meaning-by-word content assigning device, a meaning-by-meaning content giving method, and a program.

  Enormous images and videos are distributed on the Web, and if they can be associated with an existing dictionary, visual information that cannot be obtained from character string information alone can be used. For example, if an appropriate image is displayed at the same time when a word is examined, it is considered that the meaning of the word can be easily understood by anyone other than a child or a native speaker. In particular, in the case of a word having ambiguity, it is considered that it is easier to understand the difference in meaning more intuitively if an appropriate image can be presented for each meaning.

  Non-patent documents 1 to 3 disclosed as conventional techniques for assigning images to a dictionary (thesaurus). In recent years, a mechanism for sharing data and tagging (foldonomy) has been developed by a large number of users, and a large amount of images and videos to which tags have been added have been accumulated. For example, ImageNet according to Non-Patent Document 3 is a synset (group of synonyms) selected from the system of WordNet (Christine Fellbaum, editor. 1998. WordNet: An Electronic Lexical Database. MIT Press.), An existing thesaurus. ) A large amount of images. Images are collected by searching from the Web, and it is manually determined using AMT (Amazon Mechanical Turk, http://www.mturk.com/) whether or not the images are appropriate for the target synset.

Francis bond, Hitoshi Isahara, Sanae Fujita, Kiyotaka Uchimoto, Takayuki Kuribayashi, and Kyoko Kanzaki "Enhancing the Japanese WordNet", Proceedings of the 7th Workshop on Asian Language Resources, ACL-IJCNLP, 2009, p.1-8 Satoshi Fujii, Tetsuya Ishikawa, “Resolution of image ambiguity by text processing and application to encyclopedia site”, The 11th Annual Conference of the Language Processing Society of Japan (NLP-2005), 2005, p.1002-1005 Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li and Li Fei-Fei, "ImageNet: A Large-Scale Hierarchical Image Database", IEEE Conference on Computer Vision and Pattern Recognition, 2009, p.248-255

  Non-Patent Document 1 gives an image to a dictionary by assigning an image acquired from Open Clip Art Library (OCAL) to Japanese WordNet (http://nlpwww.nict.go.jp/wn-ja/). ing. Here, after comparing the OCAL and WordNet hierarchical structures to obtain image candidates, it is manually determined whether the images are appropriate for each synset (semantic class). This method has the advantage that OCAL is copyright free and can be redistributed, but because the types of images are limited, it is difficult to cover the meaning of the dictionary widely.

  In addition, the method for assigning an image to a dictionary in Non-Patent Document 2 is to associate an image acquired from the Web with the meaning of the dictionary search system CYCLONE (http://cyclone.cl.cs.titech.ac.jp/). It is. The meaning that the image should correspond to is estimated by eliminating the ambiguity of the link source text of the image. However, when there is a large difference in the appearance frequency of a plurality of meanings, only the most significant meaning images can be obtained even if only the headword is searched. For example, when searching for the entry word “arch”, there was no image indicating “home run” in at least the top 500 images. As described above, it is difficult to obtain an appropriate image even for a minor meaning in a search using only a headword.

  In addition, the method of adding an image to a dictionary in Non-Patent Document 3 is excellent in that a large amount of data can be collected with high accuracy. However, since the target synset is limited at present, the completeness of the meaning is questionable. Remains. ImageNet uses the broader terms to expand the search term and use it for the search. However, the broader word may work well, but it may obscure the meaning of the word. For example, when “boiled and dried” is expanded with the broader term “food”, many “foods using boiled and dried” appear. As described above, when only the broader word is used for the expansion, the relevance rate of the search may be lowered.

  An object of the present invention is to provide content such as images and videos suitable for each meaning when there is a definition word indicating the headword and its meaning, or a similar text, in view of the problems of the prior art. An object of the present invention is to provide a meaning-by-meaning content assigning apparatus, a meaning-by-meaning content giving method, and a program that can cover a plurality of meanings even in a minor meaning.

  The content providing apparatus classified by meaning of the present invention includes a search word selection unit and a content search unit.

  The search word selection unit inputs a list in which a plurality of search word candidates each having a relationship with the head word are classified according to the meaning of the head word, and one or more search words for each word meaning based on the relationship. Select and output.

  A content search part searches a communication network using the said search word corresponding to the said meaning, and acquires a predetermined content for every meaning of the said headword.

  According to the word meaning-specific content assigning device, word sense-specific content assigning method, and program of the present invention, when there is a headword and a definition sentence indicating the meaning or similar text, contents such as images and videos suitable for each meaning In addition, even a minor meaning can cover a plurality of meanings widely.

The block diagram which shows the structural example of the content provision apparatus 100 according to meaning. The figure which shows the example of a processing flow of the content provision apparatus 100 according to meaning. The figure which shows the example of the dictionary. The figure which extracted the part of meaning and a definition sentence from FIG. The figure which shows another example of the dictionary. The figure which extracted the part of meaning and a definition sentence from FIG. The figure which shows the example of the list | wrist which produced | generated the several search word candidate which each attached | subjected the relationship with a headword from each definition sentence of FIG. 4, distinguishing for every meaning. The figure which shows the variation of the relationship between a headword and a search word candidate. The figure which extracted only the word which has a synonym relation with headword "EU" from the explanatory note of FIG. The figure which shows the example of a screen display at the time of giving an image to the dictionary of FIG. The figure which shows the example of a screen display at the time of giving an image to the extraction information of FIG. The block diagram which shows the structural example of content provision apparatus 100 'classified by meaning. The figure which shows the example which displayed the search result containing an image content according to meaning when searching with the headword "sunflower".

  Hereinafter, embodiments of the present invention will be described in detail.

  FIG. 1 is a block diagram showing a configuration example of the meaning-by-meaning content providing apparatus 100 of the present invention, and FIG. The content assignment device 100 according to meaning of the present invention includes a search word extraction unit 110, a search word selection unit 120, and a content search unit 130. In the following description, the case where the target natural language is Japanese is taken up, but any natural language such as English, Chinese, Spanish, German, French may be used.

  The search word extraction unit 110 receives a dictionary including at least a headword and a text that expresses one or more meanings of the headword, and a plurality of words having a predetermined relationship with the headword from the text for each meaning. The search word candidates are extracted, and a list in which a plurality of search word candidates each having a relationship with the entry word are classified by meaning is generated (S1).

Examples of the dictionary 10 to be input to the semantic content granting device 100 include a basic word meaning database Lexeed (see Reference 1 and FIG. 3), a user participation dictionary on the Web (http://en.wikipedia.org/), Various things are possible, such as WordNet, a thesaurus that includes explanations of semantic classes (synsets). Moreover, even if it is not organized as a dictionary, it is also conceivable to input a word word and its description-like portion extracted from the Web.
[Reference 1] Kaname Kasahara, Hiroshi Sato, Francis Bond, Takaaki Tanaka, Sanae Fujita, Tomoko Kanesugi, Akinari Amano, "Construction of" Basic Word Meaning Database: Lexeed "", Information Processing Society of Japan Research Report, 2004, 2004-NLC -159, p75-82

  3 and 5 are diagrams showing examples of the dictionary 10. FIG. 3 shows a case where a basic word meaning database is used as the dictionary 10, in which related information relating to the three word meanings for the headword “arch” is compiled. As shown in FIG. The part of the definition sentence of three meanings can be extracted. FIG. 5 shows a case where the ambiguity avoidance page of Wikipedia (registered trademark) is used as the dictionary 10, and from this, as shown in FIG. 6, seven meaning definition parts related to the headword “EU” are extracted. be able to. From the page of FIG. 5, three, two, and two definition sentences are extracted for each of the three sub-words “EU”, “Eu”, and “eu” under the headword “EU”. You can also.

  The search word extraction unit 110 extracts a plurality of search word candidates having a predetermined relationship with the head word from the extracted head word and text such as a definition sentence and an explanatory sentence that express one or more meanings of the head word. And a list in which a plurality of search word candidates each having a relationship with a headword are classified according to meaning. 7 extracts a plurality of search word candidates having a predetermined relationship with a headword from each definition sentence in FIG. 4, and distinguishes a plurality of search word candidates each having a relationship with a headword for each meaning. It is an example of the generated list. In this list, almost all search word candidates that can be extracted are extracted from the definition sentence of FIG. 4 under the relationship of synonyms, broader terms, related words, and fields with the headword. As described above, an arbitrary relationship such as a narrow word, an entire part, an abbreviation, or another name may be set and extracted. When extracting, focusing on a specific relationship, for example, it is possible to extract only synonyms, extract only synonyms and broader terms, or extract related terms with weights. For example, FIG. 9 shows only words having a synonym relation with the headword “EU” extracted from the explanatory text of FIG. 5 (the meaning number is given for convenience). Also, if there are synonyms, synonyms will be extracted, if not, broader terms will be extracted, the relationship will be different depending on the meaning to obtain different search terms by meaning, or other definitions in the definition It is also conceivable to extract words with a low appearance frequency in sentences.

For extracting the relationship information as described above, a known method shown in Reference 2 or the like can be used. However, rules corresponding to the dictionary may be individually created and extracted. For example, in Wikipedia, in-site links are provided as shown by the underline in FIG. 5, it is also conceivable to extract such in-site links and highlighted portions of characters as related words. At this time, it is conceivable that a word linked from many pages is not extracted because it is difficult to show the characteristics of the meaning, and a link to a time expression such as an age is not extracted. If a synonym is to be extracted, the leading part before the hyphen (-) in the explanatory text of FIG. 5 can be extracted and extracted as a synonym for the headword "EU". Alternatively, when the name of the reference destination is explicitly indicated by a word or an arrow, such as “one of the twelve branches . → 戊 ”, the name of the reference destination can be extracted as a synonym. Alternatively, when the headword is alphanumeric like “computer science” for the headword “CS”, it can be extracted as a partial synonym including each character.
[Reference 2] Francis Bond, Eric Nichols, Sanae Fujita, and Takaaki Tanaka, "Acquiring an Ontology for a Fundamental Vocabulary", COLING-2004, 2004, p.1319-1325

  The search word selection unit 120 receives a list in which a plurality of search word candidates generated by the search word extraction unit 110, each of which is associated with the head word, are classified according to meaning. One or more search terms are selected and output for each meaning based on the relationship between the word candidates and the headwords (S2). For example, when a list as shown in FIGS. 7 to 9 is input, select only a synonym, or select only a broader term, or if there is a synonym, specify a synonym, and if there is no synonym, select a field. If there is no word to indicate, it is possible to select a broader word. It is also conceivable to select a plurality of types for the relationship, or to make the number of selections within a certain upper limit value. Furthermore, it may be possible not to select a related word in a minor meaning using the appearance frequency, meaning number, link frequency, etc. in a sense bank or dictionary. Further, when the meaning is very minor as the meaning of the headword, it is possible to use only a related word such as a synonym without using the headword as a search word. For example, in the case of “arch 3”, only “home run” may be used. In addition, when the headword is a foreign word, it is possible to use only the original word as a search word. For example, in the case of a foreign artist name, the spelling in the original language is used for the search instead of Japanese (Katakana etc.).

  The content search unit 130 searches the communication network using a search word corresponding to each meaning, thereby obtaining predetermined content for each word meaning of the headword (S3), and assigning or linking this to the dictionary 10 To create a dictionary 10 '. FIG. 10 shows an example of screen display when an image is added to the dictionary of FIG. 3, and FIG. 11 shows an example of screen display when an image is added to the information of FIG. Here, in addition to images, the content to be acquired may be any content desired to be added to text that displays meaning, such as a moving image. The content search can use a known image search system on the Web or any other method. Known image search systems on the Web include, for example, Google (registered trademark) (Google AJAX images API, http://code.google.com/intl/ja/apis/ajaxsearch/), goo (registered trademark) (http : //bsearch.goo.ne.jp/) and Yahoo! (registered trademark) (http://image-search.yahoo.co.jp/).

  By configuring as described above, when there is a definition sentence indicating the headword and its meaning, or a similar text, contents such as images and videos suitable for each meaning can be given, and minor meaning Even so, it can cover a wide range of meanings.

  In the case where a list as shown in FIGS. 7 to 9 in which a plurality of search word candidates each having a relationship with the headword are divided according to the meaning for headwords having one or more meanings can be prepared. The search term extraction unit 110 is not necessary. In such a case, as shown in FIG. 12, it is possible to configure the semantic content adding apparatus 100 ′ including the search word selection unit 120 and the content search unit 130 and execute only S <b> 2 and S <b> 3 in FIG. Good.

  Although FIG. 1 illustrates an example in which content is assigned to a dictionary, a list of search term candidates classified by meaning obtained in the search term selection unit 120 is stored in a database, and a search request is received from a user. It is also conceivable that the content search unit 130 performs a search by meaning in real time and displays the search results in real time by meaning. FIG. 13 is an example in which a search result including image content by meaning is displayed when a search is made with the headword “sunflower”.

  When the above-described content-by-word content granting apparatus and word-by-meaning content giving method are realized by a computer, the processing functions of the search word extraction unit, the search word selection unit, and the content search unit are described by a program. By executing this program on the personal computer or portable terminal through the exchange of data between the input means and various storage means and the CPU, the hardware and software cooperate to realize the above processing functions on the computer. The effect of the content provision apparatus classified by meaning and the content provision method classified by meaning of invention is produced. In this case, at least a part of the processing function may be realized by hardware. Further, the various processes described above are not only executed in time series according to the description, but may be executed in parallel or individually according to the processing capability of the apparatus that executes the processes or as necessary. In addition, it can change suitably in the range which does not deviate from the meaning of this invention.

Claims (5)

For a headword, a list in which a plurality of search word candidates each having a relationship with the headword are classified according to meaning is input, and one or more search words are selected and output for each meaning based on the relationship A word selector,
A content search unit that acquires a predetermined content for each meaning of the headword by searching a communication network using the search word corresponding to the meaning;
Meaning-specific content granting device comprising: In the content provision apparatus classified by meaning according to claim 1,
At least a headword and a text each representing one or more meanings of the headword are input, and for each meaning, a plurality of search word candidates having a predetermined relationship with the headword are extracted from the text, and the headword And a search word extraction unit that generates the list in which a plurality of search word candidates each having a relation to the above are classified by word meaning. The search word selection unit reads a list in which a plurality of search word candidates each having a relationship with the head word are classified according to the meaning of the head word, and selects one or more search words for each word meaning based on the relationship Search term selection step to be output,
A content search step in which a content search unit acquires a predetermined content for each meaning of the headword by searching a communication network using the search word corresponding to the meaning;
The content granting method according to meaning of meaning. In the content provision method according to meaning of claim 3,
Prior to the search word selection step,
The search word extraction unit is input with at least a headword and a text each representing one or more meanings of the headword, and for each meaning, a plurality of search word candidates having a predetermined relationship with the headword from the text. A word-by-phrase content adding device that further performs a search word extraction step of generating the list in which a plurality of search word candidates respectively extracted and associated with the headword are classified by word meaning.   A program for causing a computer to function as the apparatus according to claim 1.