JP2003242144A - Similar kanji retrieval method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To enable the retrieval of similar kanji (Chinese characters) in a short time only by input of optional kanji information by generating a retrieval vector from a plurality of pieces of kanji information inputted, and performing a similarity retrieval with a preliminarily constituted vector database to eliminate the labor and time for selecting the radical or the number of storks of a kanji character. <P>SOLUTION: For a registered kanji character obtained from an input means 1, kanji constituting element information is generated by a generation means 2, and similar kanji constituting elements for facilitating the retrieval from constituting elements are newly generated and added thereto by a generation means 3. These pieces of element information are combined together by a means 4, and a vector database mainly having components of the elements to the registered kanji character is generated and stored by a means 5. In kanji retrieval, the vectors of kanji information inputted by the input means 1 are acquired from the database by a means 7, the input vectors are combined by a means 8 to extend the retrieval information. The similarity calculation of the combined vector with the vector space of the vector database is carried out to output kanji candidates with ranking in a short time. <P>COPYRIGHT: (C)2003,JPO

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a kana-kanji conversion for inputting characters and documents in a word processor, a portable terminal, etc. The present invention relates to a similar kanji search support device capable of inputting kanji in a short time. 2. Description of the Related Art With the spread of word processors, portable terminals and the like, opportunities to input difficult-to-read Chinese characters are increasing, and the number of Chinese characters is increasing due to international communication in the Asian region. I have. Since the reading of difficult Chinese characters is often unknown, the following two types of input support methods have been put to practical use. In the first conventional method, there is a palette input method in which a radical or stroke number is input from a palette prepared by the system to narrow down and search for a corresponding kanji, but most of the radicals are not familiar and are easily input. It is difficult to count the number of strokes, and it is difficult to search for kanji in a short time. The second conventional method is a kanji recognition method based on handwriting input, which solves the drawbacks of the first conventional method, but it is difficult to input a difficult kanji with many strokes in a short time with a mouse or the like. In addition, it is difficult to use for handicapped people and the elderly. further,
There is a disadvantage that input support using voice recognition, which has been progressing in recent years, cannot be easily handled. In the document search, a method of obtaining statistical data of words and character strings existing in a document, constructing a vector for the document in which weights for those components are calculated, and searching for a similar document by the vector similarity method. Yes, the present invention combines this vector similarity technique with Kanji search. SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and eliminates the need for selecting a radical and the number of strokes in a pallet system. An object of the present invention is to solve the problem of inconsistency between constraints and speech recognition input. Furthermore, in the conventional method, accurate selection of radicals and stroke numbers and accurate input are always required, and there is no flexibility in the search information, which is inconvenient for the user. It is characterized by a device that introduces similarity and relevance in meaning and shape, and can input and specify wider search information. [0007] According to the kanji search data construction method according to the claims, search information (similar kanji, radical, reading of radical, etc.) for difficult kanji to be searched is freely input. A similar kanji search method and apparatus capable of calculating a similarity between an input vector for the search information and a vector database constructed in advance from the components of the kanji information, and outputting kanji candidates in descending order of similarity. The kanji vector and its database are expanded by newly adding related / similar configuration information to the component set obtained directly from the kanji information, and using the weight of the component using the inverse ratio of the frequency of all components. It is characterized by building. [0008] A kanji search method includes means for merging input kanji vectors for a plurality of pieces of input kanji information, and similarity search of merged vectors from a vector database for similarity search and ranking output. Features. This eliminates the need for the input user to select kanji components from the screen, etc., and simply inputs the kanji information (similar kanji, radicals, reading of radicals, etc.) that the user wants to search. You can search with The present invention will be specifically described below with reference to the drawings showing an embodiment thereof. FIG. 1 is a block diagram showing a configuration of a similarity search apparatus according to the present invention (hereinafter, referred to as the present invention). In FIG. 1, a kanji component set is generated by a component set generation unit 2 for registered kanji obtained from an input unit 1 such as a keyboard, voice input, handwriting input, file input, and range designation. For example, for the Chinese character “Hohong”,
The components of “Sanzui (radical)” and “Hong Kong” are determined, but the components of “Humi” and “Mu” are generated from “Hong Kong”. (Radical) "," mu "," bow "｝ are obtained. These constituent elements are elements for disassembling kanji parts (parts) and are not data based on a new technology, so that they can be easily constructed manually. In the present invention, new information is added to the kanji parts disassembly. For example, for the difficult kanji character “tortoise”, similar related components “turtle”, “monkey”, “turtle”, etc. are generated.
The similar kanji set generating means 3 is added as new information. The component merging means 4 merges the two element sets of the kanji parts and the new information. The vector generation storage means 5 generates and stores a weight vector for the registered kanji from the entire set of elements of all registered kanji. is there. As an example of calculating the weight of the components of this vector, the frequencies of all the components are summed from the component set of each kanji, and the components with high frequency are given a small weight, and the components with low frequency are given large weight. The calculation method is general, and the following individual kanji vector VEC (K) is generated for the kanji K. VEC (K) = (W (x1), W (x2),.
, W (xn)) where xi (0 <i <n + 1) is a component and w (xi) is a calculated weight. A set of these vectors is stored in the database as an overall vector. [0013] Kanji registration when creating a database,
Kanji is registered in ascending order of strokes, and a vector database is constructed step by step. For example, if “soil” is registered in three strokes, each component is represented by “ten” and “? (Horizontal bar)”. At this time, since “ten” has already been registered as a two-stroke kanji, the parts “?” (Horizontal bar) and “1” (vertical bar) of “ten” can be generated, and the component of the kanji “earth” is “Ten”, “? (Horizontal bar) "and" 1 (vertical bar). Next, in the same manner, the same constituent element as “Doshi” can be generated for the three strokes of Chinese character “Koji”. And can be used for automatic generation of a similar component set. In this sense, the component of “engine” is “cho”,? (Horizontal bar) ”,“ 1 (vertical bar) ”, if“ ten ”and“ cho ”are already similar components,“ soil ”is“ soil ”and“ ko ”is a similar component In this way, the similarity of the constituent elements is determined by the inclusion relation of the element set. However, according to the present invention, it is possible to search as search information regardless of whether “Soil”, “Soil”, or “Engine” is input as input information by similar elements. Information that can be expanded. In the similar kanji search device, one or more kanji similar to the kanji to be searched are input from search condition input means 11 such as a keyboard, voice recognition, range designation, etc., and respective vectors are searched from a database. A kanji vector merging unit 8 generates a retrieval vector obtained by merging the plurality of vectors (meaning that the vector is considered to be a union if the vector is considered to be a set of constituent elements). The similar kanji search means 9 performs a search by collating the database. For example, if you want to search for “Hohong” as a kanji, you can enter (“Hiro”, “Qing”), (Sanzui, Hiro),
(Bow, mu), (Hiro), etc. are conceivable. For the first input set, the main components of “Hiro” are “bow”, “mu”,
The main components of “Qing” are “Sanui,” “Blue,” “King,” and “Moon.” The similarity between the search vector and the database that combines these components is calculated. , Chinese characters with high similarity (for example,
"Elbow", "black", etc.) are output in order. As described above, the apparatus according to the present invention stores a set of kanji vectors having a component weight as a component from a set of kanji search components and their similar components in a storage device as an entire vector. In advance, the individual kanji vectors for multiple input kanji are extracted from the vector database and merged, and the merged vector is used as a search kanji vector to perform a similar search from the vector database. There is an excellent effect that similar kanji can be searched in a short time only by inputting arbitrary kanji information that can be intuitively remembered in the mind without the trouble of doing. In particular, a handicapped person or an elderly person who feels inconvenienced by handwritten kanji input becomes a better kanji input device by linking it with voice input. In addition, it is difficult for the user to change the database prepared by the system in the palette input or the handwriting input. It also has excellent properties.

[Brief description of the drawings] FIG. 1 is a block diagram illustrating a configuration of a device of the present invention. [Explanation of symbols] 1 Input means 2 Component set generation means 3 Similar component set generation means 4 Means of merging components 5 Vector generation storage means 6 Vector database 7 Kanji vector acquisition means 8 Kanji vector merging means 9 Similar search means 10 Output means 11 Search condition input means

   ────────────────────────────────────────────────── ─── Continuation of front page    (72) Inventor Kazuhiro Morita             2-3-56 Kita-Yama-cho, Tokushima-shi F term (reference) 5B009 LB03 LC01                 5B075 ND03 PP28 PR06 QM08 UU02

Claims (1)

Claims: 1. A kanji constituent element set generating means for generating a kanji constituent element set from input means such as a keyboard, voice recognition, range designation, etc., and elements related / similar to each constituent element. A similar element set generating means for generating a set, a component element merging means for merging the two sets, and a kanji vector having a component weight as a component from all the element sets for all kanji generated above. , Vector generation storage means for storing the vector database, vector acquisition means for acquiring from the vector database a plurality of kanji information vectors for the intended kanji input from the input means, and vector merging for merging the plurality of kanji information vectors Search and output kanji candidates with ranking by means and similarity calculation of merged vector and vector database Similar Kanji search method characterized by comprising a similarity search unit that.