Data processing: database and file management or data structures – Database design – Data structure types
Reexamination Certificate
1998-01-05
2001-01-16
Black, Thomas G. (Department: 2771)
Data processing: database and file management or data structures
Database design
Data structure types
C707S793000, C707S793000
Reexamination Certificate
active
06175828
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention relates to a retrieval apparatus which is applied for retrieval of information in a text database (e.g., electronic books or electronic dictionaries), video database (e.g., for photographic or video data) with an explanatory text or multimedia database (e.g., for multiplexed text data).
Information retrieval in a text database has been widely realized by using a text-based retrieval method by searching for keywords. The method is such that a key word to be searched is keyed-in at a terminal and data including the key word is searched. For an image database, retrieval of a desired image stored therein may be conducted by searching for a keyword in an explanatory text attached to the image. However, information retrieval based on keywords may fail in obtaining matched information if a keyword is entered that is semantically analogous to but does not strictly match with a target one of words in the database.
On the other hand, semantics-based retrieval may be conducted by associative retrieval using feature vectors which are context vectors proposed in a paper entitled “Associative retrieval using a large-scale document database,” TECHNICAL REPORT OF IEICE AI92-99 (1993-1) (THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS). Feature vectors used in a retrieval apparatus according to the present invention also correspond to the context vectors described above. A retrieval method using the feature vectors is disclosed in Japanese Laid-open Patent Publication (TOKKAI HEI) No. 6-195388.
The feature vector is indicative of a degree of relationship between a concept of each word in a text and a context. In other words, a degree of semantic connection between many feature words and each word in the text is expressed by vectors. If feature words are classified into N conceptual terms, each feature word corresponds to a value of each element of N-dimensional vectors. The feature vectors Xi=(xi1, xi2, . . . , xiN) of word <i> have element values xij which are not smaller than 0 and not larger than Em, i.e., 0≦xij≦Em where Em is a positive constant value. The expression xij=0 indicates that there is no relation between the word <i> and the feature word <j>. Value varies depending on the extent of relationship between them. When a feature vector consists of five feature words, e.g., [nature, city, noise, animal, green] and respective element values take either of binary values 0 and 1, a feature vector of a word, e.g., “mountain” can be expressed as (1,0,0,1,1).
As shown in
FIG. 1
, there is a prior art retrieval apparatus using feature vectors. This retrieval apparatus comprises an input for inputting a request-text for retrieval, an output for outputting a result of retrieval, a control for controlling retrieval, a word dictionary and a database. The control is used for realizing functions of word extracting structure, vector generating structure and vector retrieving structure. The word dictionary contains records of words paired with corresponding feature vectors (hereinafter called ┌word vector┘). The database holds records of data paired with corresponding feature vectors (hereinafter called ┌data vector┘). The magnitude of a data vector is normalized to have a specified constant value.
In the retrieving apparatus, retrieval starts by entering a request-text to the apparatus through the input. The word extracting structure extracts words from the request-text input through the input referring to the word dictionary. The vector generating structure reads the word dictionary to find therein a word vector corresponding to the word extracted by the word extracting structure from the request-text and converts the request-text to vectors (hereinafter called ┌request-text vectors┘) in a feature vector space. This conversion is realized by calculating a sum of word vectors of the words extracted by the word extracting structure and normalizing the obtained sum of the vectors to have a certain constant value. The vector retrieving structure calculates a distance between the request-text vectors and the data vectors in the database and the output outputs a retrieval result in the form of data arranged in the ascending order from the data having the shortest distance.
The above-mentioned retrieval method by using feature vectors, however, is based only on a distance between a request-text vector and a data vector and, therefore, can not always retrieve first data including words contained in the request-text as a keyword-retrieval method. This retrieval method can not retrieve data having a keyword or can not conduct semantic retrieval in parallel with searching a match with a keyword.
On the other hand, Japan Laid-open Patent Publication (TOKKAI HEI) No. 6-208588 discloses such a keyword-retrieval method which determines an importance degree of a keyword (depending upon parts of speech) by morphological analysis of a request-text and which allows a retrieval system to retrieve in the case that the request-text does not completely meet with the keyword in the database. This method, however, is based on keywords and, when retrieving a word or words separated in a request-text, can not retrieve data composed of only word or words that are semantically close to but not included in the request-text.
SUMMARY OF THE INVENTION
The present invention has as its object the provision of a retrieval apparatus which is capable of conducting both retrievals by using keyword and feature vector at a time, determining the order of possibility of the retrieval results according to successful combination of both retrievals and outputting improved retrieval results.
Another object of the present invention is to provide a retrieval apparatus, which includes an input for inputting a request-text for requesting retrieval of a target, a word dictionary for storing words together with corresponding word vectors, a database for storing data, each including at least a word, together with corresponding data vectors, word extracting structure for extracting a word in the request-text by using the word dictionary, vector generating structure for generating a request-text vector from the request-text by using the word dictionary and the word extracting structure, vector retrieving structure for determining semantic similarity of each of the data vectors to the request-text vector, key-word retrieving structure for retrieving the data including a word being common to that included in the request-text by using the word extracting structure and extended retrieving structure for retrieving data adapted to the request-text on the basis of the semantic analogy of data vectors to the request-text vector and the correspondence of the word in the data to the word in the request-text, which is detected by the keyword retrieving structure.
A further object of the present invention is to provide a retrieval apparatus, which includes an input for inputting a text requesting retrieval of a target, a word dictionary for storing words together with corresponding word vectors, a database for storing data including at least a word, word extracting structure for extracting a word in the request-text by using the word dictionary, vector generating structure for generating a request-text vector from the request-text by using the word dictionary and the word extracting structure and for generating a data vector from the data, vector retrieving structure for determining semantic similarity of each of the data vectors to the request vector, keyword retrieving structure for retrieving the data including a word being common to that included in the request-text by using the word extracting structure and extended retrieving structure for retrieving data adapted to the request-text on the basis of the semantic similarity of the data vector to the request-text vector and the correspondence of the word in the data to the word in the request-text, which is detected by the keyword retrieving structure.
A s
Ikeuchi Hiroshi
Keshi Ikuo
Kuromusha Kenichi
Black Thomas G.
Mills John G.
Nixon & Vanderhye P.C.
Sharp Kabushiki Kaisha
LandOfFree
Retrieval apparatus does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Retrieval apparatus, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Retrieval apparatus will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2540452