JP4129132B2 - Search result presentation apparatus, search result presentation method, and search result presentation program - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a search request statement generation device, a search result presentation device, a search request statement generation method, a search result presentation method, a search request statement generation program, a search result presentation program, a data search device, a data search method, and a data search program. For example, the present invention relates to an image information search.
[0002]
[Prior art]
Various methods for searching image data such as images and videos have been proposed, and one of them is a search using impression expression words as a key.
For example, there is a data search device of Japanese Patent Application No. 11-281201 (unpublished) proposed by the present applicant. This search device is based on a technique of inputting impression expressions, degree adverbs, and objective characteristic conditions as search keys by text when searching for image information.
[0003]
[Problems to be solved by the invention]
However, this method includes the following problems.
First, it is difficult to create an appropriate search request sentence in the following cases.
For example, if the user does not grasp a clear subjective request that he / she desires, and if he / she can imagine his / her request only vaguely, it is difficult to express the search request by sentences.
[0004]
In addition, when the user cannot think of what subjective characteristics the subject matter in the content has, that is, when the user has not obtained detailed information about the target content, You may not know what subjective image the thing you have can have.
For example, in wine content, a user who is not familiar with liquors such as wine may not come up with an impression word such as “sweet” or “fruity” which is an index for selecting wine.
Further, for a user who is familiar with keyword searches such as a search site on the Internet, words related to feeling may be difficult to come up even though a noun phrase is conceived.
[0005]
In addition to this, for example, a problem also arises when the noun phrase (evaluation viewpoint) on which the impression expression word is applied is unclear. That is, when only the impression expression word is used as a search key, the user may implicitly recognize the object of the impression expression word, and the system may not be able to produce an appropriate result for the user's intention.
For example, as shown in FIG. 25, in the hotel content, even if the user implicitly recognizes and requests that “the room is large” as a search request, the system side does not intend the user only from “wide”. Even sentences such as “wide bathroom” and “wide entrance” are acquired.
[0006]
Further, for example, the data search device of Japanese Patent Application No. 11-281201 outputs the search results in the order of image information having higher similarity, but the display of the search results also includes the following problems.
First of all, it was difficult to know which part was hit. Unlike keyword search, data containing impression expression words with similar elements is also hit, so it is difficult to determine which word caused the hit only by highlighting where the character faces match.
[0007]
Also, the search results are difficult to understand at first glance, and the user's satisfaction was low. That is, even if attention is paid to the matching of the character faces, it is not possible to grasp each feature or difference, so the sense of satisfaction is lowered.
In addition, if the result of the search does not match the user's intention or there is no applicable result, a re-search is necessary, but when re-inputting, the same result is displayed and the relevant result is displayed. Search request sentences are difficult to come up with and are a burden on the user.
[0008]
Therefore, an object of the present invention is to provide a search result presentation device, a search result presentation method, a search result presentation program, and the like that can clarify the match between a search request and a result.
[0009]
[Means for Solving the Problems]
In order to achieve the above object, according to the first aspect of the present invention, search result acquisition means for acquiring search data assigned a search order, and document data included in the search data are acquired by the search result acquisition means. Document data acquisition means, extraction means for extracting impression expression words and noun phrases linked to the impression expression words for each search data from the document data acquired by the document data acquisition means, and extraction means Search using the control unit for controlling the impression expression word and the noun phrase extracted in the above, and the set of the impression expression word and the noun phrase controlled by the control unit. The impression expression word and the previous expression excluding the impression expression word and the noun phrase set included in the search data having the first search order from the set of the impression expression word and the noun phrase included in the data Difference means for acquiring a set of noun phrases for each search data of the second or lower search order, and the impression expression word and the noun phrase acquired by the difference means for the search data of the second or lower search order A search result presentation device characterized by comprising: a presenting means for presenting a set of
In the invention described in claim 2, when there is a degree adverb associated with the impression expression word and the noun phrase, the extraction means extracts the degree adverb as a set. A search result presentation device is provided.
In the invention according to claim 3, the emphasis presenting means for emphasizing and presenting at least one of the degree adverb, the impression expression word, or the noun phrase associated with the impression expression word included in the document data, Among the document data, a summary presentation means for presenting a summary of the document data using at least one of a degree adverb, the impression expression word, or a noun phrase linked to the impression expression word, The search result presentation device according to claim 1, further comprising at least one.
In the invention according to claim 4, the degree adverb, the impression expression word, or the impression expression word presented by at least one presentation means of the presentation means, the emphasis presentation means, or the summary presentation means The search result presentation device according to claim 3, wherein a search request sentence can be created using a noun phrase linked to.
According to a fifth aspect of the present invention, a search result is presented in a search result presentation device comprising a search result acquisition means, a document data acquisition means, an extraction means, a control means, a difference means, and a presentation means. A search result presenting method used when the search result acquisition means includes a first step in which the search result acquisition means acquires search data to which a search order is assigned, and the document data acquisition means in the first step. A second step of acquiring the document data included in the search data, and the extraction means is linked to the impression expression word and the impression expression word for each search data from the document data acquired in the second step. A third step of extracting a noun phrase as a set; a fourth step in which the control means controls the impression expression word and the noun phrase extracted in the third step; However, using the impression expression word and the noun phrase controlled in the fourth step, the search is performed from the set of the impression expression word and the noun phrase included in the search data of which the search order is second or lower. A set of the impression expression word and the noun phrase excluding the set of the impression expression word and the noun phrase included in the search data having the first rank is acquired for each search data of the second or lower search rank. And a sixth step in which the presenting means presents the set of the impression expression word and the noun phrase acquired in the fifth step with respect to the search data of the second or lower search rank. Provided is a search result presentation method characterized by comprising:
In the invention according to claim 6, a search result acquisition function for acquiring search data assigned a search order, a document data acquisition function for acquiring document data included in search data by the search result acquisition function, An extraction function for extracting a set of impression expression words and noun phrases linked to the impression expression words for each search data from the document data acquired by the document data acquisition function, and the impression expression words extracted by the extraction function And the control function for controlling the noun phrase, and the impression expression word included in the search data in which the search rank is second or lower using the impression expression word and the noun phrase set controlled by the control function And the set of the noun phrase and the noun phrase excluding the set of the impression expression word and the noun phrase included in the search data having the first search rank from the set of noun phrases A difference function that is acquired for each search data that is ranked second or lower, and a presentation function that presents the set of impression expression words and noun phrases acquired by the difference function with respect to search data that is second or lower in the search order And a search result presentation program for realizing the above on a computer.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
(First embodiment)
Hereinafter, a first embodiment of the present invention will be described in detail with reference to FIGS. FIG. 1 is a diagram showing a configuration of a data search apparatus 11 equipped with a search request sentence generation system and a search result presentation system according to the first embodiment. The data search device 11 performs image information search as an example, but is not limited to this. For example, the data search device 11 is widely used in search devices using impression expression words (sensitivity words) such as “wide” and “sweet”. Generally applicable.
The data search device 11 includes a central control unit 12, an input / output unit 16, a modem 17 storage device 18, and the like.
The input / output unit 16 is connected to an input device such as a keyboard, a microphone, and a mouse, an output device such as a display device, a printer, and a speaker, or an input / output device such as a magnetic disk drive. Then, the user can input a search request sentence via the input device or obtain a search result via the output device. Further, data can be written to and read from a magnetic disk or the like using a disk drive or the like.
[0011]
The storage unit 18 is a storage device configured by, for example, a hard disk or other nonvolatile memory. The storage unit 18 includes a search request generation program, a search result display program, a search program such as an image information search program, a program unit 19 in which other programs are stored, and various data such as content data in a database. The data section 20 that is used, the notation fluctuation control dictionary 24 that is used when controlling the use of impression expressions (adjectives, etc.), which will be described later, and the evaluation viewpoint (noun phrases strongly linked to the impression expressions) are controlled. It includes a thesaurus dictionary 25 for use in the process.
[0012]
The modem 17 is a part for the data search device 11 to communicate with the outside via a network such as the Internet.
The central controller 12 includes a CPU (Central Processing Unit) 13, a ROM (Read Only Memory) 14, and a RAM (Random Access Memory) 15.
The RAM 15 is a readable / writable memory that loads a program of the program unit 19 and provides a working memory to the CPU 13.
[0013]
The ROM 14 is a read-only memory that stores basic programs and parameters for operating the data search apparatus 11.
The CPU 13 performs various calculations and determinations according to a program stored in the ROM 14 or a program loaded in the RAM 15, and transmits / receives data to / from the input / output unit 16, the modem 17, the storage device 18 and the like via the bus line 21 , And control of these devices.
[0014]
Of the search request sentence generation system and the search result display system, first, the search request sentence generation system will be described.
FIG. 2 is a block diagram showing the configuration of the search request sentence generation system 37.
Among the constituent elements of the search request sentence generation system 37, the notation fluctuation control dictionary 24 and the thesaurus dictionary 25 are stored in the storage device 18 (FIG. 1), and the merge module 32, the control module 33, the noun phrase frequency module 34, The sort module 35 and the query tool display module 36 are modules implemented by software by loading the search request statement generation program stored in the program unit 19 into the RAM 15.
The input means 31 is, for example, a keyboard or mouse connected to the input / output unit 16, and the user can input search conditions and the like using the keyboard and mouse.
[0015]
Next, the function of the search request sentence generation system 37 will be described with reference to FIGS.
When the merge module 32 acquires the input data from the input means, it merges all the data in the content, and takes out a set of impression expressions and noun phrases that are strongly associated with the impression expressions. For example, when degree adverbs are associated with impression expressions such as “slightly broad” or “pretty sweet”, these degree adverbs are extracted together with the impression expressions and noun phrases.
When creating a query tree to be described later, it is not necessary to use an adverb to the extent that is associated with an impression expression word. deep.
Here, the impression expression word is a noun phrase or a single adjective or adjective verb, such as "wide" in "wide room" or "fruity" in "fruity wine" It is a phrase that expresses the impression and sensibility.
[0016]
FIG. 3 is a diagram for explaining the function of the merge module 32. First, merge data 41 is generated by merging all the data data1, data2,. Next, an impression expression word and a noun phrase strongly associated with the impression expression word are extracted from the merge data 41 as a set, and impression expression word noun phrase set data 42 is generated.
These noun phrases mean impression objects and are also called evaluation viewpoints.
In the example of FIG. 3, the impression expression word “wide” and the noun phrase “room” linked thereto are extracted as a set from “room is wide” in the merge data 41 and registered in the impression expression word noun phrase set data 42. Yes. Similarly, “in a quiet environment” to “quiet” / “environment” set, and “room is wide” to “room is” / “wide” set are taken out.
[0017]
Next, the control module 33 (FIG. 2) controls the impression expression word and the noun phrase extracted in the impression expression word noun phrase set data 42.
First, regarding the control of impression expressions (adjectives, adjective verbs, etc.), if these adjectives and adjective verbs are inflected forms, they are controlled to the final form. In addition, referring to the notation fluctuation dictionary 24, a standard that has been determined in advance, such as Kanji, Hiragana, or Katakana whose readings are the same and synonymous with any normalized expression (for example, hiragana, katakana notation is aligned with Kanji notation). Control).
Next, the control module 33 refers to the thesaurus dictionary 25 and performs notation control for the noun phrases in the impression expression noun phrase set data 42. When there is a broader word in a noun phrase, this notation control is aligned with the broader word, and if there is a synonym, it is regulated to a normalized expression.
[0018]
FIG. 4 is a diagram for explaining the function of the control module 33.
First, the control module 33 acquires the impression expression word noun phrase set data 42, and although not shown, for example, the impression expression words such as “wide” and “quiet” are terminated with “wide”, “quiet”, etc. To control.
Next, the control module 33 controls the impression notation fluctuation using the notation fluctuation dictionary 24. For example, “Hiroi” and “Hiroi” are “wide”, “Sweet” is “sweet”, “Smell”, “Kusai”, “Kusai” are “smell”, etc. Controlled by normalized expressions.
In this way, the notation fluctuation dictionary 24 has notations that need to be controlled and notations (normalized expressions) after they are controlled as a database.
[0019]
Next, the control module 33 uses the thesaurus dictionary 25 to, for example, “bathroom” as a synonym “bath”, “big bath” as a broader term “bath”, and “sento”. Is controlled to the broader word “bath”, and post-notation controlled data 43 is generated. In the post-notation controlled data 43, the set of impression expression words and noun phrases has the same notation such as “wide” / “room” and “wide” / “room”, for example.
The thesaurus dictionary 25 is a database that stores inferior words, corresponding broader words, synonyms that should be aligned, synonyms after being aligned, and the like.
[0020]
The noun phrase frequency module 34 (FIG. 2) obtains post-control data from the control module 33, sorts only the noun phrases in the data according to the frequency, and sorts the frequent ones with an evaluation viewpoint (impression expression words). It is adopted as an object of impression, for example, “house” of “large house”.
The noun phrase related to the impression expression word existing in the content is the subject of subjective expression such as “wide” or “thick” in the content, and the user selects the subject of the subjective expression. It can be said that this is the point of view that is desired as a search request.
[0021]
FIG. 5 is a diagram showing the sort data 45 generated by the sort module 35. The sort data 45 indicates the frequency of noun phrases extracted from the post-notation-controlled data 43 by the sort module 35. This frequency is calculated regardless of the impression expression word to which the noun phrase is linked.
From FIG. 5, the frequency of “room” is 25 (that is, the number of times “room” appears in the noun phrase in the post-notation control data 43), and “bath” is 13 or the like.
The sort module 35 generates the viewpoint data 46 together with the expression word from the frequency obtained from the sort data 45 with the noun phrase having the highest frequency as the viewpoint. In the sort data 45 of FIG. 5, “room”, “bath”, “lobby”, and “employee” having a frequency of 10 or more are selected as evaluation viewpoints and output from the sort module 35 as viewpoint data.
[0022]
FIG. 6 is a diagram showing the viewpoint data 46 generated by the sort module 45. In the viewpoint data 46, the noun phrase selected as the evaluation viewpoint and the impression expression word associated with the noun phrase are associated with the frequency.
The query tree display module 36 obtains the viewpoint data 46 from the sorting module 35, sets the evaluation viewpoint as a parent (node), and on the other hand, inputs / outputs 16 as a tree structure (query tree) having a set of impression expression words as children. Display on the display device connected to 1).
In addition, the user can use the query tree when searching. That is, when the user selects an impression expression word and a noun phrase from the query tree by a mouse operation or the like, the query tree display module 36 generates a search request sentence in which the selected impression expression word and noun phrase are set, and the data Send to search system.
[0023]
FIG. 7 shows an input screen 48 displayed on the display device. The input screen 48 is displayed by an image information search program or the like. The input screen 48 includes a feeling input field 49, a text input field 50, an objective information input field 51, a search button 52, and the like.
The query tree displayed in the feeling input field 49 is generated and displayed by the query tree module 36 of the search request statement generation program.
[0024]
As shown in FIG. 7, in the feeling input field 49, the evaluation viewpoint (noun phrase) is a node, and impression expressions connected to the evaluation viewpoint are visually displayed as a tree structure. That is, the noun phrase is hierarchically displayed with the noun phrase as the parent and the impression expression word linked thereto as the child. When multiple impression expressions are associated with the same noun phrase, these impression expressions are siblings.
In this way, by displaying a high-frequency noun phrase and impression expression word in a tree structure, for example, even a user who does not recall the impression expression word “fruity” that is usually used in wine is “wine The appropriate impression expression word can be easily obtained from the query tree, such as “-fruity”.
The query tree can be used as a clue when the user sets the evaluation viewpoint when the evaluation viewpoint of the user is ambiguous.
[0025]
Next, how to use the input screen 48 will be described.
When the user uses the feeling input field 49, the user selects and determines the evaluation viewpoint in the feeling input field 49 by clicking a node by operating the mouse. Then, impression expression words corresponding to the evaluation viewpoint are displayed in a tree structure.
In the feeling input field 49 of FIG. 7, impression expressions such as “wide”, “clean”, and the like are displayed in the tree as a result of the user clicking the evaluation viewpoint “room”, “bath”, “location”. It is displayed by the structure.
[0026]
For example, when the user clicks the evaluation viewpoint “room” and the impression expression word “wide” and then clicks the search button 52, image information having a set of “wide” / “room” in the image information attached document is searched. The
FIG. 8 shows a case where search information is input to the text input field 50. The text input field 50 is a text search request text input frame. A set of impression expression words and noun phrases is taken out from the sentence inputted from here, and if there is a degree adverb, it is also taken out as a set.
For example, although there is no evaluation point of “meal” in the feeling input field 49, when “delicious meal” or the like is entered in the sentence input field 50 in a natural sentence, “meal” − “delicious” in the feeling input field 49. You can get the same results as selecting this tree.
[0027]
In the text input field 50, a keyword can be searched by inputting a specific data name.
The data search apparatus 11 includes a keyword search system (not shown) and performs a search using this.
FIG. 9 shows a specific data name entered in the text input field 50.
For example, the user can input a specific name “Shinagawa Hotel” in the text input field 50.
FIG. 10 is a search result display screen, which is a result of a search performed by a keyword search system (not shown) displayed on a display device by a search result display system which will be described later.
[0028]
Here, the user knows that he / she was drawn by either or both of the elements “close to the station” or “new room”, selects the impression expression phrase that he / she needs and executes a search. be able to.
Further, by clicking an appropriate attribute in the objective information input field 51 (for example, “Aomori Prefecture”), objective information can be used in a restrictive manner, and the width of the result display can be adjusted.
The impression expression phrase is composed of three elements: a degree adverb, an impression expression word representing an impression assumed by the degree, and a noun phrase strongly associated with the impression expression word.
[0029]
Next, the search result display system will be described.
The functions of the search result display system 61 are as follows.
When displaying the text of the text attached to the image information, the degree adverb, impression expression word, noun phrase, etc. are highlighted with color coding etc. to differentiate from other parts of the text. When displaying a summary of image information, a sentence having many impression phrases is displayed in a predetermined number of lines, for example, one line.
Further, for the data of the second rank or lower, the difference information from the first rank is acquired and displayed.
[0030]
FIG. 11 is a block diagram showing the configuration of the search result display system 61.
The notation fluctuation control dictionary 24 and the thesaurus dictionary 25 are the same as those used in the search input section sentence generation system 37.
The search information is processed in order from the bottom to the top in FIG. First, information retrieved by an image information retrieval program or the like is input to the impression expression noun phrase adverb phrase set module 62.
In the following, description will be made assuming that search ranks of the first and second ranks are displayed as an example, but this may display more search results.
[0031]
The impression expression noun phrase adverb phrase set module 62 extracts the attached explanatory text of the image information of the first rank and the image information of the second rank. It is assumed that the search results acquired by the search result display system 61 from the data search device are ranked in advance.
Then, an impression expression word, a set of impression expression words, noun phrases, and degree adverbs are extracted from the attached explanatory text extracted in the same manner as the search request sentence generation system 37.
Next, the control module 63 controls the notation of the set of impression expression words and noun phrases using the thesaurus dictionary 25 and the notation fluctuation control dictionary 24. Note that the control module 63 of the search request sentence generation system 37 can also be used as the control module 63.
For impression expressions (adjectives, adjective verbs, etc.), if it is an inflection form, it will be a final form. Also, kanji, hiragana, and katakana, which have the same reading and are synonymous, are controlled to any normalized expression.
For noun phrases, control to broader terms, and if there are synonyms, control to normalized expressions.
[0032]
The difference extraction module 64 compares the set of noun phrases and the post-control impression expression words extracted from the two documents, and extracts only the different parts. This reveals the common and different aspects of both image information. Then, the difference extraction module 64 displays an impression expression phrase different from the first-order image information for the second-order or lower image information.
Note that the difference extraction module 64 compares the set of controlled impression expressions and noun phrases.
The emphasis module 65 differentiates from other words / phrases by color-coding the degree adverbs such as “youthful” and “spicy”, impression expression words, noun phrases, and the like.
[0033]
FIG. 12A is a diagram illustrating a state in which the search result display system 61 performs data processing using wine content as an example.
First, the impression expression word noun phrase adverb phrase set module 62 obtains document data 71 attached to the image information of the search result, and then extracts a set of impression expression words and noun phrases from each data to generate set data 72. To do.
For example, a set of “fresh” / “taste” is extracted from “fresh taste” of the document data 71. If there is an adverb such as “Very”, this is also taken out as a set.
[0034]
Next, the control module 63 controls impression expression words and noun phrases. For example, notation such as “refreshing” / “taste” is controlled by normalized expressions such as “refreshing” / “taste”. The control data 73 is obtained.
Next, the difference extraction module examines a set different from the common set (impression expression word and noun phrase) of the two data, and extracts the difference data 74.
Thus, for example, the first search result object (hereinafter referred to as wine 1) and the second search result object (hereinafter referred to as wine 2) are both “fresh” / “taste”. Compared to wine 2, it has “very”, “sweet” / “fragrance”, and wine 2 has “beautiful” / “label” and “elegant” / “fragrance” compared to wine 1. I understand that.
[0035]
Furthermore, it is possible to display the search result as shown in FIG. 12B using the difference information.
FIG. 12B shows the search result displayed on the display screen.
The difference between the wine “Cuvée Cruise Rouge” ranked first and the wine ranked second to fourth is shown for each wine.
As described above, when the difference information is extracted with respect to the result of the second place or less, the user can more clearly recognize the characteristics of each wine.
[0036]
FIG. 13 is a diagram showing an input screen 48 when searching for wine content as an example.
Here, when the user selects “taste” / “light” from the tree structure and selects “do not select” as the objective information, the search result screen 77 shown in FIG. 14 is displayed.
First, in the text display field 78, the adverb module 65 such as “youthful”, “spicy”, “fruity”, “light and fresh”, impression expressions, noun phrases, etc. are colored and displayed by the highlighting module 65 to be highlighted. ing.
In the summary display column 79, a portion having a lot of impression phrases is displayed for the number of lines designated in advance (here, two lines).
[0037]
In the first search result display column 80, a set of impression expression words, noun phrases and degree adverbs extracted from the image information attached document is displayed, and the characteristics of this wine are displayed in an easy-to-understand manner. In the second search result display column 81, the difference from the first wine is displayed.
[0038]
The three types of result display methods, emphasis of degree adverbs by color coding, impression expression words, summary display, and difference display, have been shown above. If the user selects an impression expression phrase from each of the second-ranked search result display fields 81 with a mouse click or the like, the search can be performed again using the impression expression phrase as a search request sentence.
As a result, even if the user makes a search once and sees actual data to clarify his / her request, the search can be executed again by selecting an impression expression phrase including the request.
For example, if “fresh” / “feel” matches the user's request more than “taste” / “light” input first, for example, the first search result display field 80 The user can search again by selecting “fresh taste” by clicking with the mouse.
[0039]
FIG. 15 is a flowchart for explaining the operations of the search statement generation system 37 and the search result display system 61.
First, the query generation system 37 displays a query tree with the evaluation viewpoint as a node in the feeling input field 49. Then, the user selects a search request sentence by selecting an evaluation viewpoint and an impression expression word (step 10).
Next, for example, a search is performed using an expression word and an evaluation viewpoint selected by a search system such as an image information search system (step 20).
[0040]
Next, the impression expression word noun phrase adverb phrase set module 62 of the search result display system 61 extracts document data attached to the searched image information and the like of the first and second ranks, and the impression expression word from the document data And a set of noun phrases and degree adverb phrases (step 30).
Next, the control module 63 controls the usage form of the impression expression word to the end form (step 40), and further controls the impression expression word (step 50) and the noun phrase (step 60).
Next, the difference extraction module 64 acquires the degree of duplication of the set of impression expression words, noun phrases, and degree adverb phrases (step 70), and compares the first and second rank data to extract differences ( Step 80).
Next, the search result is displayed on the display device (step 90).
Further, if re-search is necessary, after selecting the impression expression phrase necessary for re-search from the difference information (step 100), the search execution button 52 is selected (step 110).
[0041]
The following effects can be obtained from the search result display system 61 described above.
First, when presenting the search results, extract the sentence that contains more impression expression phrases from the accompanying explanatory text displayed as a result, and show the summary effect by showing the specified number of high-order sentences Can do.
In addition, each element of the impression expression phrase (degree adverb, impression expression word, noun phrase) in the attached explanatory text displayed as a result can be easily seen by highlighting or the like.
Further, in the attached explanatory text displayed as a result, the impression expression phrase of the first place and the impression expression phrase of the second place or less can be displayed with the impression expression phrase of the difference from the first place.
As for the re-search, the summary part displayed in the text display column 78, the summary column 79, the first search result display column 80, the second search result display column 81 of the search result screen 77, or the degree adverb, When any one of the impression expression word, the emphasized part of the noun phrase, or the difference is selected, the search can be performed again using this as a query.
[0042]
(Second Embodiment)
By the way, the present applicant has proposed a data search device of Japanese Patent Application No. 11-281201 (unpublished). In the second embodiment, the data search device is used as the data search system described in the first embodiment.
Japanese Patent Application No. 11-281201 is not configured to search for image information using a set of impression expression words and noun phrases, but can be searched using a set of impression expression words and noun phrases. Partially change the system.
First, the data search device of Japanese Patent Application No. 11-281201 will be described.
[0043]
The data retrieval apparatus includes a subjective evaluation information dictionary in which impression data for image information to be searched is placed in a coordinate space having a subjective similarity element as a coordinate axis, and a coordinate space having a subjective similarity element as a coordinate axis. A subjective evaluation expression dictionary in which impression data for an impression expression word serving as a search key is laid out, a search condition sentence acquisition means for acquiring a search condition sentence subjectively expressed in a natural language as a search condition, and the search condition A search impression that extracts an impression expression word from a search condition sentence acquired by a sentence acquisition unit and generates impression data for the acquired search condition sentence from impression data of the impression expression word stored in the subjective evaluation expression dictionary Image information of impression data most similar to the impression data generated by the data generation means and the search impression data generation means Find an image information retrieval unit which is provided with output means for outputting the image information retrieved by the image information retrieval unit, a (first configuration).
In the first configuration, impression data for each impression expression word stored in the subjective evaluation expression dictionary includes range data for designating a range for each coordinate axis (second configuration).
In the first configuration or the second configuration, a text data acquisition unit that acquires text data relating to new image information, an impression expression word is extracted from the text data acquired by the text data acquisition unit, and the subjective evaluation is performed. Impression data generating means for generating impression data for the new image information from the impression data of the impression expression word stored in the expression dictionary, and the impression data generated by the impression data generating means is used as the impression of the new image information. Image information adding means for registering it in the subjective assessment information dictionary as data (third configuration).
In addition, the first means, the second means, or the third means are registered in the subjective evaluation expression dictionary in the extraction means for extracting the text data impression expression words and the impression expression words extracted by the extraction means. Impression expression word impression data generating means for generating impression data for the unregistered impression expression word from impression data for the registered impression expression word when there is an impression expression word and an unregistered impression expression word, and the impression Impression expression word adding means for registering impression data generated by the expression word impression data generating means in the subjective evaluation expression dictionary as impression data of the unregistered impression expression words can be configured (first operation). 4 configuration).
Furthermore, the subjective evaluation information dictionary of any one of the first configuration to the fourth configuration includes an objective feature condition for each image information, and the image information search means includes the search condition When an objective feature condition is acquired from the acquisition means, the image information can be narrowed down using the objective feature condition as a constraint condition (fifth configuration).
Furthermore, the subjective evaluation information dictionary of any one of the first to fifth configurations includes objective feature conditions together with impression data for each image information, and provides objective new image information. Based on the feature condition, impression data for the new image information is generated from the relationship between the impression data of the image information stored in the subjective evaluation information dictionary and the objective feature information, and stored in the subjective information dictionary. A second image information adding unit can be provided (sixth configuration).
Further, the search condition acquisition unit of any one of the first configuration to the sixth configuration acquires the search condition from an external device connected to a network, and the output unit includes the output unit It can be configured to output image information to an external device connected to the network (seventh configuration).
Further, as a data search program related to the data search device, a function for creating a subjective evaluation information dictionary in which impression data for image information to be searched is placed in a coordinate space having a subjective similarity element as a coordinate axis; , A function that creates a subjective evaluation expression dictionary in which impression data for impression expression words serving as search keys is placed in a coordinate space with subjective similarity elements as coordinate axes, and is expressed subjectively in natural language as a search condition A search condition sentence acquisition function for acquiring the search condition sentence, and impression data of the impression expression word extracted from the search condition sentence acquired by the search condition sentence acquisition function and stored in the subjective evaluation expression dictionary A search impression data generation function for generating impression data for the acquired search condition sentence, and a mark generated by the search impression data generation means. An image information search function for searching image information of impression data most similar to data from the subjective evaluation information dictionary, and an output function for outputting image information searched by the image information search means are realized by a computer. There are a data search program and a recording medium on which the data search program is recorded.
[0044]
(1) Outline of data retrieval device
A subjective evaluation information dictionary 5 (see FIG. 21) in which impression data for image information to be searched is placed in a coordinate space having subjective similarity elements as coordinate axes. Further, a subjective evaluation expression dictionary 4 (see FIG. 18) in which impression data of impression expression words serving as search keys is placed in a coordinate space having a subjective similarity element as a coordinate axis. The impression data of the subjective evaluation expression dictionary 4 includes coordinate values for each seat axis and range data for designating a range on each coordinate axis around the coordinate value (see FIG. 19).
Then, as a search condition, an impression expression word is extracted by natural language processing from a search condition sentence subjectively expressed in natural language, and corresponding impression data is extracted from the subjective evaluation expression dictionary 4. When there are a plurality of impression expression words, one or a plurality of impression data integrated by the integration process is generated as search impression data. Then, in the coordinate space having the subjective similarity element as the coordinate axis, the image information having the impression data having the closest Euclidean distance to each search impression data is output as the search result.
On the other hand, when adding new image information, an impression expression word is extracted from text data such as an explanatory text added to the added image information, and integrated into one impression data by integration processing. The impression data consisting of the coordinate values obtained by removing the range data from the impression data is registered in the subjective evaluation information dictionary 5 as impression data of image information.
In addition, when an unregistered impression expression word is extracted in the subjective evaluation expression dictionary 4 in the process of image information search and image information addition, other impression expression words included in sentences and text data used in the process are extracted. One impression data is generated by integrating the impression data, and is registered in the subjective evaluation expression dictionary 4 as impression data of the impression expression word.
[0045]
(2) Details of data retrieval device
FIG. 16 conceptually shows the configuration of the data search apparatus.
As shown in the conceptual block diagram, the data search apparatus includes an input unit 1, a search condition sentence acquisition unit 2, a search impression data generation unit 3, a subjective evaluation expression dictionary 4, a subjective evaluation information dictionary 5, and an image information search unit 6. Output means 7, image information adding means 8, and impression expression word adding means 9.
The input means 1 is for inputting a search condition sentence in a natural language desired by the user, and a keyboard or a voice recognition device is used.
The search condition sentence acquisition unit 2 acquires a search condition sentence input from the input unit 1 or acquires a search condition sentence transmitted from a wired connection or a wirelessly connected external device using a communication unit. Examples of external devices include other personal computers connected via LAN, and external personal computers connected to the Internet via the World Wide Web (WWW). The search condition statement can be acquired.
The search impression data generation means extracts impression expression words and the like from the search condition sentences acquired by the search condition sentence acquisition means, extracts impression data corresponding to the extracted impression expression words from the subjective evaluation expression dictionary 4, and searches impressions. Generate data.
[0046]
The subjective evaluation expression dictionary 4 classifies, for example, impression expression words that are used when expressing impressions such as images and feelings about a desired beer in a natural language when searching for beer. This is a stored dictionary. The impression expression word here is not a physical feature of the image information to be searched but a word (sensitivity word) that expresses an intuitive impression with adjectives and adjective verbs such as “soft” and “clear”. is there.
Here, all the data formats that can express the relative relationship by subjective similarity can be handled as image information to be searched. For example, in addition to images such as videos and still images, music data, audio data, Text data itself such as publications, other signal data, and contents having these as elements are also included.
FIG. 17 conceptually shows the contents of the subjective evaluation expression dictionary 4.
As shown in FIG. 17, the subjective evaluation expression dictionary is classified into various categories for image information to be searched such as beer, wine, sake, movies, TV program information, travel information on the Web, online shopping, and the like. Has been. And impression expression words used to express the impression of image information belonging to each category, specifically, expression words represented by adjectives and adjective verbs and degree adverbs (adverbs indicating the degree of “somewhat”, “pretty”, etc.) In addition, nouns, proper nouns, and the like that are strongly associated with impression factors are collected in advance for each category and stored in the subjective evaluation expression dictionary.
[0047]
On the other hand, for each category, an element (subjective similarity element) representing subjective similarity is extracted with respect to image information belonging to the category. For example, when the category of the target image information is a movie, subjective similarity elements such as “thrill”, “feeling”, and “comedy” are extracted as the properties of the work. For the category “beer”, as shown in FIG. 17, subjective similarity elements such as “koku”, “crime”, “umami”, “nodomi”, and “exhilaration” are extracted.
Then, by specifying the strength of the subjective similarity element for each impression expression word, the impression data for each impression expression word (a multidimensional, which is placed in a coordinate space with the subjective similarity element as a coordinate axis) Coordinate values (vectors) have been determined, i.e., for collected impression expressions, each expression specifies the strength of a subjective similarity element, i.e. points to a specific range in the impression data space. Based on the hypothesis, the relative level value of each similarity element is given.
For example, as shown in FIG. 17, the impression expression word “with richness” for the category “beer” has a relative value of 3 for the subjective similarity element “rich” and a relative value for the subjective similarity element “Kire”. Impression data with values of 0, ... are given.
[0048]
FIG. 18 exemplifies a coordinate space having a subjective similarity element as a coordinate axis, and impression data for each impression expression word placed in the coordinate space. In FIG. 18, the coordinate space of beer is expressed using the three elements of subjective similarity elements of richness, sharpness, and umami as coordinate axes, and impression data for impression expressions “creamy”, “mellow”, and “dry” are placed. It is a thing.
As shown in FIG. 18, each impression expression word is expressed by impression data (vector) having a subjective similarity element extracted for image information belonging to the category as a coordinate axis and each coordinate value as an element. Is done.
[0049]
As illustrated in FIG. 17, the actual impression expression word includes range data indicating the range included in the impression expression word in addition to the coordinate value indicating the relative value of each subjective similarity element. Yes. Therefore, the impression expression word is specified in an impression data space surrounded by a range centered on the coordinate value of each subjective similarity element (a constant range centered on the coordinate value).
FIG. 19 shows an impression data space of impression expression words in two-dimensional coordinates. As illustrated in FIG. 19, for example, the impression expression word “creamy” of the category “beer” has a range of plus or minus 2 centered on the coordinate value 4 with respect to the subjective similarity element “koku”, That is, it is specified in the range of 2 to 6, and is specified in the range of plus or minus 2 around the coordinate value −2, that is, in the range of −4 to 0 with respect to the subjective similarity element “clear”.
Thus, the impression data of the impression expression word has a certain range with respect to each coordinate axis (subjective similarity element), so that it is not represented by a point on the coordinate space, but a certain space on the coordinate space. The image information included in the impression data space of the impression expression word can be listed as search candidate image information. By listing image information in the impression data space of impression expression words in this way, it is not necessary to determine the degree of similarity with impression data for all image information, and the similarity with impression data for the listed image information Judgment can be made, and the judgment can be speeded up.
[0050]
The subjective evaluation information dictionary 5 stores data related to search objects, and impression data and attached information for each image information are classified and stored for each category. The category in the subjective evaluation information dictionary 5 is the same as the category in the subjective evaluation expression dictionary 4, and the search is performed within the range of the same category.
FIG. 20 conceptually shows the contents of the subjective evaluation information dictionary 5.
As illustrated in FIG. 20, impression data for each image information is represented by a vector whose elements are values on the coordinate axis composed of common subjective similarity elements, similar to impression data for impression expression words. . That is, the impression data of the image information “Asahi Super Dry” is a vector (3, 4, 2, 3.5, 3. 5, ...).
Similarly, “Asahi Kurose” “Asahi First Lady” “Asahi Super Premium” “Asahi Draft Beer Dunk” (both are registered trademarks or trademarks of Asahi Breweries Co., Ltd.), “Kirin Lager Beer” "Beer""Kirin Ichiban Kuro draft beer""Kirin Beer Factory <Raw""" Kirin Blau Meister "(both are registered trademarks or trademarks of Soba Sake Co., Ltd.) It is expressed by
Note that each coordinate value of the impression data for the image information is not attached with range data unlike the impression expression word.
Therefore, image information is laid out in a multidimensional coordinate space with subjective similarity elements as coordinate axes, and by specifying an area (impression data space) having a certain width in the coordinate space, the impression data space It is possible to search (list up) image information placed inside.
On the other hand, as the attached information, objective feature conditions such as numerical data, target image data, text data such as explanatory text, and the like are stored for each image information. In the attached information, various data that cause impression of other image information are stored as related information in various forms.
[0051]
The objective feature condition stored in the subjective evaluation information dictionary 5 as the attached information means physical feature data of the image information, and is quantitative data that can be measured quantitatively. In addition, category information obtained by classifying image information according to its properties and uses also corresponds to objective feature conditions.
For example, when the image is image information, the color and shape are applicable, and when the passenger car is image information, the displacement, fuel consumption, weight, and the like are applicable.
The objective feature condition is used to narrow down image information as a constraint condition when searching for image information.
[0052]
Various types of data that can impress image information include personal names, product names, etc., which have a certain kind of strong impression, or impression data of primitive figures that are components of images To do. For example, when making a movie work image information, the impression of the leading actor may be more dominant than the impression of the work itself. In such a case, an impression expression word extracted from the commentary of the movie work, In some cases, the impression data given to the leading actor, etc. is quoted rather than the impression data estimated from the scene image, and the impression data of the work is more reflective of subjective similarity. The name of the leading actor is stored as attached information.
[0053]
As shown in FIGS. 17 and 20, the subjective evaluation expression dictionary 4 and the subjective evaluation information dictionary store category classification information such as nouns included in each category. For example, “beer” is stored as category classification information for determining the category “beer”, and “wine” is stored as category classification information for the category “wine”. In addition to storing each category's noun itself as categorization information, for example, for the category “wine”, related words such as “Bordeaux” and “Beaujolais Nouveau”, “(grape + grape + grape) × brewing” And the like are also stored as category classification words. Related words and conditional expressions are also stored as category classification information for other categories.
In this manner, by storing the category classification information in the subjective evaluation expression dictionary 4, nouns that match (or satisfy the conditional expression) are included in the text data in the search process, the image information addition process, and the impression expression word addition process. If included, it can be automatically classified into the corresponding category.
[0054]
The image information search means 6 (FIG. 16) uses the search impression data (a set of level values of subjective similarity elements) generated by the search impression data generation means and an objective feature condition as a search key. Search various image information stored in.
That is, the image information registered in the subjective evaluation information dictionary 5 is narrowed down by using objective feature conditions as constraints, and image information within the range indicated by the level value set is extracted from the image information. The similarity of impression data is judged and output as search results in the order of image information with higher similarity.
[0055]
The output unit 7 outputs the search result searched by the image information search unit 6. The output of the search result by the output means 7 is performed by outputting the searched image information to the display device and printing it out to the printing device.
[0056]
The image information adding means 8 generates impression data for the new image information and stores it in the subjective evaluation information dictionary 5 together with the attached information. Impression data of new image information is generated by extracting impression expression words stored in the subjective evaluation expression dictionary 4 from text data attached to the image information and using the impression data of the impression expression words. Also. The impression data of the registered image information is generated from the objective characteristic condition of the image information registered in the subjective evaluation information dictionary 5 and the objective characteristic condition for the new image information.
In this way, the image information adding means 8 can newly add image information from text data or objective characteristic conditions, so that the subjective evaluation information dictionary can be easily enriched.
[0057]
The impression expression word adding means 9 generates impression data for an unregistered impression expression word and additionally registers it in the subjective evaluation expression dictionary 4. A new impression expression word is added by the impression expression word adding means 9 in a text in a natural language inputted when searching for image information, or an impression expression word not registered in the image information adding means 8 is a text corresponding to the image information. This is done when there is an unregistered impression expression word in the data.
Thus, since the new impression expression word is automatically registered in the subjective evaluation expression dictionary 4 by the impression expression word adding means 9, the more the data retrieval device is used, the more the subjective expression expression dictionary 4 becomes. Become. In particular, the user's feelings and sensitivities to the image information, as well as the texts that express them, are diverse. By collecting and adding new impression expressions in the search process, it is possible to increase the sensitivity and expression of more users. The corresponding subjective evaluation expression dictionary 4 can be obtained.
[0058]
By the way, the subjective evaluation expression dictionary 4 and the subjective evaluation information dictionary 5 of the data search device of Japanese Patent Application No. 11-281201 (unpublished) related to the present applicant described above store impression expression words and noun phrases as a set. As a result, there was a problem that the evaluation viewpoint was not determined. That is, for example, when a user creates a search sentence such as “wide room” and searches for it, information such as “wide entrance” and “wide bathroom” is also hit because the evaluation viewpoint is not determined.
Therefore, in the second embodiment, subjective similar elements are generated using a set of impression expression words and noun phrases.
[0059]
The subjective evaluation expression dictionary 4 (hereinafter simply referred to as the expression dictionary 4) of the data search apparatus (hereinafter referred to as the previous data search apparatus) of Japanese Patent Application No. 11-281201 is, for example, savory and crisp with respect to beer. Although impression data is set for a rich, creamy, impression expression word, the expression dictionary 4 of the second embodiment has, for example, an impression expression word “wide” as shown in FIG. ”Includes a set of impression expression words and noun phrases such as“ large room ”and“ wide bathroom ”. That is, the subjective similarity element is composed of a set of impression expression words and noun phrases. A set of impression expression words and noun phrases such as “wide bathroom” is a subjective feature element, but takes a value of 5 ± 3. This corresponds to the expression dictionary 4 of the previous data retrieval apparatus in which the characteristic feature element has a range of “3 ± 2” as “rich”.
[0060]
On the other hand, the subjective evaluation information dictionary (hereinafter simply referred to as the information dictionary 5) in the previous data retrieval apparatus has subjective similarity elements created for each image information, for example, “full”, “clear”, “umami” Impression data is created for impression expression words such as "". On the other hand, as shown in FIG. 24, the information dictionary 5 according to the second embodiment is a set of impression expression words and noun phrases such that the subjective similarity element is “bathroom is wide” or “room is wide”. Impression data has been created.
As described above, in the second embodiment, the vector space can be configured by subjective similarity elements having evaluation viewpoints such as “room is wide” and “bathroom is wide”.
In other words, in the information dictionary 5 using the example of FIG. 18, a space is created by subjective similarity elements such as “mellow”, “rich”, and “dry”. In the second embodiment, Furthermore, the evaluation viewpoint is further clarified, and a space is created by a set of impression expression words and noun phrases such as “mellow smell”, “mellow taste”, “rich taste”, and so on.
[0061]
As described above, in the previous data retrieval apparatus, the subjective similarity element was determined without simply setting the evaluation viewpoint such as “wide”. However, in the second embodiment, the evaluation viewpoint is further clarified. Since the subjective similarity element is composed of impression expression words and noun phrases such as “bathroom is large” and “room is large”, the accuracy of search can be improved.
That is, when the user creates a search request sentence with “a hotel in a large room”, the image information having a value of 5 ± 3 (integrated value) in the component of the subjective similarity element “room is wide” is hit. However, this is not the case, and the image information in which the component “wide bathroom” is 5 ± 3 does not hit.
[0062]
As described above, in the second embodiment, the subjective similarity element of the previous data retrieval apparatus is composed of impression expression words, and is a set of impression expression words and noun phrases. . Other search principles are the same as the previous data search apparatus.
[0063]
FIG. 22 is a diagram showing a state where the search request sentence generation system 37 (FIG. 2) and the search result display system 61 (FIG. 11) described in the first embodiment are incorporated in the previous data search apparatus. . Note that the previous data search apparatus can handle subjective similarity elements by a set of impression expression words and noun phrases.
The search request sentence generation system 37 generates a search sentence from the information acquired from the input unit 1 and outputs it to the search condition sentence acquisition unit 2. In this search sentence, an impression expression word and a noun phrase are set, and if there is a degree adverb, this is also taken out as a set.
On the other hand, the search result display system 61 receives the search result from the image information search means 6 and outputs it to the output means 7.
[0064]
In the second embodiment, the search request sentence generation system 37 and the search result display system 61 can be used for the data search device, and data search using impression expression words can be performed while setting the evaluation viewpoint.
[0065]
(Third embodiment)
In the second embodiment, subjective similarity elements (coordinate axes) are defined using a set of impression expression words and noun phrases (modified noun phrases) strongly associated with the impression expression words, and image search is thereby performed. However, this system had the following unresolved issues.
[0066]
(1) The impression expression word in the search request sentence has no modified noun phrase.
For example, when an image search is performed on hotel content, if a user inputs a query only as “wide”, the data search system according to the second embodiment assumes that there is no evaluation viewpoint, and the expression “wide room”. Even if there is ancillary data (data in the content) of the image information describing (the viewpoint of “room” is specified), it does not hit.
However, in hotel content, “wide” often expresses “room”, and from the search request sentence “wide”, the attached data with the description “room is wide” is also searched. Should.
[0067]
For example, if “wide” is simply entered as a search request sentence, a search key “wide” is generated, and accessory data describing expressions such as “very wide” and “anyway wide!” Are searched. However, attached data that is described as “very large room” or “single room is large!” Is not searched.
In addition, if “wide” is entered as a search request, even if the evaluation viewpoint “room” is automatically added, such as “room is large”, for example, content related to hotel contents or transportation charges, etc. It is necessary to set an evaluation viewpoint (noun phrase) for each content. Entrusting this to the user is a burden on the user, and a certain amount of prior knowledge is required for setting, so there may be a user who feels difficult.
[0068]
(2) The impression expression word in the attached data does not have a noun phrase strongly associated with it.
For example, if the search request sentence is “room is wide”, “room” is acquired as the evaluation viewpoint, and only “wide” is described in the attached data, and data without the evaluation viewpoint is not searched.
For example, if you enter “Search for a hotel with a large room” as a search request statement, the search key “Room is large” is generated, and “This hotel has a large room” or “Wide room” is described. Attached data is searched. However, the attached data described as “Very wide” or “Anyway here is wide” is not searched. For this reason, when setting the subjective similarity element (coordinate axis) of each attached data, the information that the description “wide” may be handled in the same way as “the room is wide” is required.
However, this is difficult to set in advance because the evaluation viewpoint changes depending on the content.
[0069]
(3) In the case where an adverse effect occurs by dividing the viewpoint in the search request sentence and the attached data.
For example, in a hotel content, “cheap price” and “cheap room” are often synonymous. However, since it is considered that “price” and “room” are not associated with each other in the existing thesaurus, in many cases “cheap price” and “cheap room” are interpreted as different meanings.
For example, if you enter "I want to find a cheap hotel" as a search request, the search key "Cheap price" will be generated, such as "This hotel is cheap" or "I can stay for a low fee" Attached data described as follows is retrieved. However, attached data such as “this hotel's room is cheap” or “anyway here is cheap!” Is not searched.
[0070]
Furthermore, this problem has different aspects depending on the content.
For example, in hotel content, “cheap” refers to “room charges”, but in content related to transportation charges, distinguish between “buses” or “trains” (evaluation perspectives). Is important.
It is a considerable burden on the user to leave the evaluation point to the user for each expression word.
For this reason, it is necessary to determine whether or not it is necessary to divide the evaluation viewpoints both when dealing with a search request sentence and when setting a subjective similarity element (coordinate axis).
This determination is made when there is no modified noun phrase in the impression expression word in the search request sentence of (1) above, and the impression expression word in the attached data of (2) has a noun phrase strongly associated with this. If not, it is necessary as well.
[0071]
Therefore, in this embodiment, the following four points are targeted.
(1) The evaluation viewpoint is automatically determined when there is no noun phrase strongly associated with the impression expression word in the search request sentence.
(2) When there is no noun phrase strongly associated with the impression expression word in the attached data, the evaluation viewpoint is automatically determined when the subjective similarity element (coordinate axis) is set.
(3) Automatically distinguishing search request sentences that do not require evaluation viewpoints.
(4) Automatically determine data that does not require separate evaluation viewpoints in the attached data.
[0072]
Hereinafter, the present embodiment will be described.
First, in order to achieve the objectives (1) and (2), in the present embodiment, an impression expression word having no strongly connected noun phrase is used by using a flag setting table 111 (FIG. 33) described later. Automatically set the evaluation viewpoint.
First, the flag setting system 100 that generates the flag setting table 111 will be described. The flag setting system 100 can be implemented in software by the CPU 13 executing a flag setting program stored in the program unit 19 (FIG. 1).
[0073]
FIG. 26 is a diagram illustrating an example of the configuration of the flag setting system 100.
The flag setting system 100 includes a merge module 90, an impression expression / noun phrase set extraction module 91, a control module A 92, an evaluation viewpoint selection module 93, a control module B, a noun phrase pickup module 95, a calculation module 96, and a most frequent noun phrase module. 97, a flag setting module 98, and the like.
[0074]
The merge module 90 merges all the attached data (all data in the content) to generate merge data.
[0075]
The impression expression word / noun phrase set extraction module 91 extracts an impression expression word and a noun phrase strongly associated with it from the generated merge data.
FIG. 27 is a diagram illustrating an example of a set of merge data, extracted impression expression words, and noun phrases.
The figure shows, as an example, impression expressions such as “Wide / Room”, “Quiet / Environment”, etc., from merged data obtained by merging Data 1, Data 2,. And an extracted set of noun phrases.
In this embodiment, it is assumed that processing is performed for each content such as hotel content, wine content,.
[0076]
The control module A92 performs notation control on the impression expression word extracted by the impression expression word / noun phrase set extraction module 91. For example, if the impression expression word is an adjective or an adjective verb inflection form, the notation control is used as an end form. A certain kanji, hiragana, or katakana is controlled to a normalized expression.
[0077]
The evaluation viewpoint selection module 93 focuses only on the noun phrase in the set of impression expression words and noun phrases extracted by the impression expression word / noun phrase set extraction module 91, and extracts the appearance frequency for each noun phrase. Then, a noun phrase whose appearance frequency is equal to or higher than a preset threshold value is selected as an evaluation viewpoint when the impression expression word is not accompanied by the modified noun phrase.
That is, if there is no noun phrase strongly associated with the impression expression word, any of the noun phrases selected here is added.
[0078]
FIG. 28 shows an example in which the noun phrases included in the impression expression word / noun phrase set extracted by the impression expression word / noun phrase set extraction module 91 are sorted in the order of appearance frequency.
As shown in the figure, the appearance frequency of the modified noun phrase “room” included in the set of impression expression words and noun phrases is 50, and is ranked at the top. Subsequently, the appearance frequency of “hotel” is 43, the appearance frequency of “feel” is 15,.
And in this Embodiment, the threshold value of appearance frequency was set to 8, and the noun phrase of appearance frequency more than this was selected as an evaluation viewpoint.
[0079]
The control module B94 refers to the thesaurus dictionary, controls the viewpoint determined by the evaluation viewpoint selection module 93 as the highest level, and controls the broader terms within that range. In addition, if there is a synonym dictionary, the thesaurus dictionary is regulated to normalized notation.
Noun phrases controlled by broader terms and normalized notation form a noun group. For example, “bath”, “sento” and “bathroom” controlled by “bath” form a “bath” group.
[0080]
FIG. 29 is a diagram for explaining control of a set of impression expression words and noun phrases.
The extracted impression expression word and noun phrase set 101 includes notations such as “wide / room”, “quiet / room”, or “large public bath” (not shown), and includes notations and subordinate words.
First, the notation shaking control dictionary 102 controls notation shaking such as “Hiroi” → “wide”, “sweet” → “sweet”, “crispy” → “odor”.
[0081]
Next, the thesaurus dictionary 103 controls lower words such as “large bath” and “sento” as broader words such as “bath”, and has a synonym such as “bathroom”. For example, by controlling to “bath”, control to normalized notation is performed to obtain a set 104 of impression expression words and noun phrases after control.
[0082]
The noun phrase pickup module 95 picks up the noun phrase related to the impression expression word existing in the content for each impression expression word, and rearranges the noun phrases in descending order of frequency. At this time, the appearance frequency of impression expression words not strongly associated with the noun phrase is also counted.
[0083]
FIG. 30 is a diagram showing an example of picking up noun phrases related to the impression expression word and rearranging them in descending frequency order for each impression expression word.
For example, the impression expression word “wide” has the highest appearance frequency of 52 although there is no noun phrase to which the impression expression word strongly relates. “Room” is 49, “Bath” is 19, and so on.
[0084]
The calculation module 96 calculates the following expression for each impression expression word and each modified noun phrase.
[Frequency of modified noun phrase N attached to impression expression T] / ([Frequency of impression expression T] − [Frequency when there is no modified noun phrase]) (1)
A noun phrase having a larger calculated value of Expression (1) has a higher frequency of being associated with the impression expression word.
[0085]
FIG. 31 is a diagram illustrating an example of the calculation result of the expression (1) for each impression expression word.
The calculation result of the noun phrase expression (1) relating to each impression expression word is shown.
For example, in the case of the impression expression word “wide”, the calculation result of the modified noun “room” is 0.333, which is the largest value, followed by “bath” 0.129, “furniture” 0.034,.・ Etc.
From these tables, for example, the noun “room” is frequently associated with the impression expression word “wide”, and the noun “transport” is frequently associated with the impression expression word “convenient”.
[0086]
The most frequent noun phrase module 97 extracts only the most frequently occurring noun from the calculation result of the calculation module 96 for each impression expression word.
Next, a predetermined threshold is set for the calculated value, and an impression expression word having a calculated value equal to or greater than the threshold is specified.
If there are no nouns that are strongly associated with impression expressions in the search request section, or if there are no nouns that are strongly associated with impression expressions in the attached data, and the calculated value for these impression expressions is greater than the previous threshold. Is larger, the most frequently used noun phrase obtained by the calculation module 96 is assigned as a noun phrase (evaluation viewpoint) strongly associated with the impression expression word.
[0087]
A table 110 in FIG. 32 shows the most frequently used nouns arranged in descending order of numerical values of the calculation results for each impression expression word.
As shown in the figure, the largest numerical value of the calculation result is the modified noun “room” associated with the impression expression word “beautiful”, and the numerical value of the calculation result is 0.458. “Price” that leads to “cheap”, “room” that leads to “clean”, and so on. In this embodiment, as an example, 0.100 is set as the threshold value of the calculation result.
For example, if the impression expression word “beautiful” appears in the search request without accompanying the modified noun phrase or appears in the attached data, the modified noun that is set as “clean” in FIG. "Room" will be linked as an evaluation viewpoint.
[0088]
The flag setting module 98 determines that, for each coordinate axis of the subjective similarity element (coordinate axis), an impression expression word having a value on the coordinate axis has a modified noun phrase belonging to each noun phrase group set in step 5. A flag is used to determine whether the value of the coordinate axis is valid, or if there is no noun phrase that is strongly connected, whether the value of the coordinate axis is valid.
For “no noun”, a flag is set corresponding to the case where the calculation result is equal to or greater than the threshold set in the table of FIG.
[0089]
FIG. 33 is a diagram showing an example of the flag setting table 111 indicating whether or not the value of each coordinate axis is valid. However, only some of the various coordinate axes are shown.
In the table, T indicates that the flag is set to enable the coordinate axis, and F indicates that the coordinate axis is disabled without setting the flag.
[0090]
For example, if a set of impression expression words and noun phrases is acquired and the noun phrase belongs to a room group, the “room size” coordinate axis and the “cleanliness” coordinate axis are enabled and the other coordinate axes are disabled. To do.
If the noun phrase belongs to the bath group, the “bath openness” coordinate axis is enabled and the other coordinate axes are disabled.
[0091]
When an impression expression word that does not have a strongly connected noun phrase is acquired, the flag of the coordinate axis specified by the impression expression word or the coordinate axis specified by the impression expression word and the noun phrase having a calculated value equal to or greater than the threshold in Table 101 Stand up.
For example, the flag setting table 111 includes a “room size” coordinate axis, a “bath size” coordinate axis, a “low” coordinate axis, a “convenience” coordinate axis, and a “cleanliness” coordinate axis. Among these, the “room size” coordinate axis is flagged because the impression expression word and noun phrase set “wide / room” is greater than or equal to the threshold in Table 110.
Further, the “cheap” coordinate axis and the “cleanliness” coordinate axis are flagged because the impression expression words “cheap” and “clean” are equal to or greater than the threshold in Table 110.
[0092]
The flags in the flag setting table 111 function as follows.
(1) A flag function for automatically setting an evaluation viewpoint in a search request statement.
If an impression expression word “wide” that does not have a strongly connected noun phrase (evaluation viewpoint) is input as a query, a flag is set for the “room size” coordinate axis in the flag setting table 111. On the other hand, the “bath width” coordinate axis is invalid because the flag is not set. Therefore, the evaluation viewpoint “room” can be automatically set for the search request.
This makes it possible to automatically determine that the query “wide” represents searching for a value having the “room size” coordinate axis.
[0093]
(2) A flag function when setting the subjective similarity element (coordinate axis) of the attached data.
For example, when an impression expression word “wide” that does not have a noun phrase that is strongly connected appears in the attached data, this attached data is assigned a value to the “room is wide” coordinate axis flagged in the flag setting table 111. Can have, but cannot have a value on a "bath wide" coordinate axis that is not flagged.
[0094]
In this way, the flag setting table 111 can be regarded as a database storing a set of preset impression expression words and noun phrases. That is, when there is an impression expression word that does not have a noun phrase to be linked to in the query or the text data included in the attached data, an appropriate noun phrase (evaluation viewpoint) is added to the impression expression word by the flag setting table 111. Can do.
[0095]
Next, in order to achieve the objectives (3) and (4), it is necessary to divide evaluation viewpoints for impression expression words having no strongly connected noun phrases using a flag setting table (FIG. 35) described later. Determine presence or absence.
Each table shown in FIG. 30 is acquired using the merge module 90 to the noun phrase pickup module 95 described above.
Next, for each impression expression word, a value of (frequency with no modified noun phrase) / (appearance frequency of the impression expression word) is calculated. The larger the calculated value, the higher the frequency that the impression expression word is used alone without a strongly associated noun phrase.
[0096]
The fact that the impression expression word is frequently used alone means that the meaning of only the word can be expressed, and it is considered that there is no need to separate the evaluation viewpoints.
In other words, nominal phrases in a set of impression expression words and noun phrases belong to any noun group, all the coordinate axes related to the impression expression words should have values, or there is no noun that is not strongly connected. In the case of an expression word as well, all coordinate axes related to the impression expression word may have values.
[0097]
FIG. 34 shows a table 120 in which the value of (frequency without modified noun phrase) / (appearance frequency of the impression expression word) is calculated for each impression expression word, and these impression expression words are arranged in descending order of the calculation result. It is shown.
As shown in Table 120, impression expressions such as “clean”, “convenient”, and “cheap” are frequently used alone.
In the present embodiment, 0.4 is set as a threshold value, and an impression expression word having a larger calculation result value than this value is not divided into evaluation viewpoints.
[0098]
FIG. 35 is a diagram showing a flag setting table 121 that indicates whether or not the value of each coordinate axis is valid. However, only some of the various coordinate axes are shown.
In the table, T indicates that the flag is set to enable the coordinate axis, and F indicates that the coordinate axis is disabled without setting the flag.
[0099]
From Table 120, the calculation results of impression expression words “clean”, “convenient”, and “cheap” are larger than the threshold value. For these impression expression words, it is necessary to select a coordinate axis (setting an evaluation viewpoint) using a noun group. For these impression expressions, flags are set for all noun groups and nouns.
Regarding the “room width” coordinate axis and the “bath width” coordinate axis, since the impression expression word “width” in the table 120 is smaller than the threshold value, it is necessary to select the coordinate axes based on the evaluation viewpoint. Therefore, a flag is set for the noun phrase of the room group for the “room size” coordinate axis, and a flag is set for the noun phrase of the bath group for the “bath width” coordinate axis.
[0100]
The flags set by the flag setting table 121 function as follows.
(1) A flag function for automatically setting an evaluation viewpoint in a search request statement.
For example, when a query containing the impression expression word “cheap” is entered in the search request sentence, it is a value in the “cheap” coordinate axis, regardless of what noun phrase it is strongly associated with, or even if there is no strongly associated noun phrase It is possible to automatically determine that a search is made for items having.
(2) Role of flag function when setting subjective similarity elements (coordinate axes) of attached data
Similarly, the "cheap" coordinate axis has a value on the "cheap" coordinate axis, regardless of what noun phrase the word "cheap" is strongly associated with, or no noun phrase is strongly associated It becomes possible.
[0101]
Thus, the flag setting table 121 can be viewed as a database storing a set of preset impression expression words and noun phrases. That is, when there is an impression expression word that does not have a noun phrase to be linked to in the query or the text data included in the attached data, an appropriate noun phrase (evaluation viewpoint) is added to the impression expression word by the flag setting table 111. Can do. Further, according to the flag setting table, it is possible to determine whether or not to associate a noun phrase with an impression expression word. For example, it can be determined that a noun phrase is associated with “width” and a noun phrase is not associated with “cheap”.
[0102]
Both of the flag setting tables 111 and 121 described above can be used as follows in the data search apparatus (FIG. 16) of the second embodiment.
When an impression expression word that is not associated with a noun phrase is included in the search request sentence in the natural language input by the user to the input means 1, the flag setting tables 111 and 121 are referred to and a predetermined noun phrase is added to the impression expression word. The impression expression word and the noun phrase set can be generated. A search request sentence can be generated from the set of impression expression words and noun phrases, and image data can be searched.
In the second embodiment, when there is an impression expression word that is not associated with a noun phrase in the search request sentence, the evaluation viewpoint (noun phrase associated with the impression expression word) is not determined, and it is difficult to perform a search. However, in the present embodiment, it is possible to automatically supplement the evaluation viewpoint using the flag setting tables 111 and 121 and perform a search.
In particular, when the flag setting table 121 is used, when an impression expression word that is not associated with a noun phrase is included in the search request sentence, it is determined whether to associate the noun phrase with the impression expression word and set an evaluation viewpoint. Can do.
Furthermore, in this embodiment, by controlling the noun phrase linked to the impression expression word in the attached data to the optimum thesaurus (evaluation viewpoint automatically set in the search request sentence) for each content, more appropriate search can be performed. Can be done.
[0103]
The image information search means 6 can determine which coordinate axis is valid using the flag setting tables 111 and 121.
For example, it is assumed that there is attached data having a “width” coordinate axis in the subjective evaluation information dictionary 5 (FIG. 16) storing hotel content. When a search request sentence generated using the impression expression word and noun phrase set “room size” is input, the image information search means 6 makes this coordinate axis effective by enabling the “width” coordinate axis. You can also search for attached data that has a value in. This is equivalent to giving the evaluation viewpoint by adding the noun phrase “room” to the “width” coordinate axis.
In the second embodiment, when there is an impression expression word that is not linked to a noun phrase in the attached data, it is difficult to search the attached data by a search request sentence in which an evaluation viewpoint is determined. In this embodiment, the flag setting tables 111 and 121 can be used to automatically supplement the evaluation viewpoint with an impression expression word that is not accompanied by a noun phrase in the attached data and perform a search.
In particular, when the flag setting table 121 is used, it is possible to determine whether or not to add a noun phrase to the coordinate axis for each impression expression word.
Furthermore, in this embodiment, the evaluation viewpoint automatically set to the noun phrase associated with the impression expression word in the search request sentence and the impression expression word not associated with the noun phrase in the attached data is controlled to the optimum thesaurus for each content. This makes it possible to perform a more appropriate search.
[0104]
Note that the use of the flag setting tables 111 and 121 is not limited to the data search apparatus of the second embodiment, and can be widely used for data search.
That is, a set of impression expression words and noun phrases is generated by adding a predetermined noun phrase to an impression expression word that is not linked to a noun phrase in the search request sentence using the flag setting tables 111 and 121, and using this When there is an impression expression word that is not connected to the noun phrase in the content to be searched or is searched, the flag setting tables 111 and 121 are used to connect an appropriate noun phrase to the impression expression word, You can search for it.
[0105]
As described above, in the third embodiment, all the modified noun phrases (noun phrases strongly associated with some impression expression word) in the corresponding content are acquired and the frequency information is acquired. And the to-be-modified noun phrase more than the preset threshold value is set as the viewpoint. Thereby, it is possible to realize a function of automatically setting a necessary viewpoint in the corresponding content.
In addition, the thesaurus is controlled based on the viewpoint determined by the above method when controlling the noun phrases by the thesaurus. That is, the standard is the highest hierarchy, and if there are noun phrases that are lower than those standards, the standard noun is controlled. This makes it possible to perform optimal thesaurus control for each content.
Furthermore, in order to appropriately and automatically determine impression expression words that have no noun phrases and impression expression words that do not need to have different viewpoints, the frequency of noun phrases that are strongly associated with an impression expression word is extracted.
And in order to implement | achieve the process when there is no noun phrase and when it is not necessary to divide a viewpoint, the flag which can set a several viewpoint instead of setting only one viewpoint to one coordinate axis was set.
[0106]
【The invention's effect】
According to the present invention, it is possible to provide a search result presentation device, a search result presentation method, a search result presentation program, and the like that can clarify a match between a search request and a result.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a data search device according to a first embodiment.
FIG. 2 is a block diagram showing a search request sentence generation system.
FIG. 3 is a diagram for explaining functions of a merge module;
FIG. 4 is a diagram for explaining functions of a control module;
FIG. 5 is a diagram showing sort data generated by a sort module.
FIG. 6 is a diagram showing viewpoint data generated by a sort module.
FIG. 7 is a diagram showing a search screen displayed on the display device.
FIG. 8 shows a case where search information is input in a text input field.
FIG. 9 shows a specific data name entered in the text input field.
FIG. 10 is a diagram showing a search result display screen.
FIG. 11 is a block diagram showing a configuration of a search result display system.
FIG. 12 is a diagram showing how the search result display system processes data.
FIG. 13 is a diagram showing a search input screen when searching for wine content as an example.
FIG. 14 shows a search result screen.
FIG. 15 is a flowchart for explaining operations of a search statement generation system and a search result display system.
FIG. 16 is a block diagram showing a configuration of a data search device that is a source of incorporating a subjective feature element generation system.
FIG. 17 conceptually shows the contents of a subjective evaluation expression dictionary in the data search apparatus.
FIG. 18 is an explanatory diagram exemplifying between a coordinate space having a subjective similarity element as a coordinate axis and a conceptual state in which impression data for each impression expression word is placed in the coordinate space in the data search apparatus;
FIG. 19 is an explanatory diagram showing an impression data space of impression expression words in the data search device as two-dimensional coordinates.
FIG. 20 conceptually shows the contents of a subjective evaluation information dictionary in the data search apparatus.
FIG. 21 is an explanatory diagram illustrating a coordinate space having subjective similarity elements as coordinate axes and a conceptual state in which impression data for each piece of image information is placed in the coordinate space in the data search apparatus.
FIG. 22 is a block diagram showing a state in which a search request statement generation system and a search result display system are incorporated in the data search device.
FIG. 23 is a diagram showing an example of extension of the subjective evaluation expression dictionary.
FIG. 24 is a diagram showing an example of extension of the subjective evaluation information dictionary.
FIG. 25 is a diagram showing a search example in a conventional search device.
FIG. 26 is a diagram showing an example of the configuration of a flag setting system.
FIG. 27 is a diagram showing a set of merge data, extracted impression expression words, and noun phrases.
FIG. 28 shows an example of noun phrases included in a set of impression expression words and noun phrases sorted in order of appearance frequency.
FIG. 29 is a diagram for explaining control of a set of impression expression words and noun phrases.
FIG. 30 is a diagram showing an example of picking up noun phrases related to the impression expression word and rearranging them in descending order of frequency for each impression expression word.
FIG. 31 is a diagram illustrating an example of a calculation result of Expression (1) for each impression expression word.
FIG. 32 shows the most frequently used nouns arranged in descending order of numerical values of calculation results for each impression expression word.
FIG. 33 is a diagram showing a flag setting table showing whether or not the value of each axis is valid.
FIG. 34 shows a table in which the value of (frequency with no modified noun phrase) / (appearance frequency of the impression expression word) is calculated for each impression expression word, and these impression expression words are arranged in descending order of the calculation results. It is a thing.
FIG. 35 is a diagram showing a flag setting table indicating whether or not the value of each coordinate axis is valid.
[Explanation of symbols]
1 Input means
2 Search condition sentence acquisition means
3 Search impression data generation means
4 Subjective evaluation expression dictionary
5 Subjective information dictionary
6 Image information retrieval means
7 Output means
8 Image information addition means
9. Impression expression addition method
11 Data retrieval device
12 Central processing unit
13 CPU
14 ROM
15 RAM
16 Input / output section
17 Modem
18 Storage device
19 Program Department
20 Data part
21 Bus line
24 Notation shaking control dictionary
25 Thesaurus
31 Input means
32 Merge Module
33 Control module
34 Noun phrase frequency module
35 Sort module
36 Query tree display module
37 Search request sentence generation system
41 Merge data
42 Impression expression noun phrase set data
43 Post-regulation data
45 Sort data
46 Viewpoint data
48 Input screen
49 Feeling input field
50 Text entry field
51 Objective information entry field
52 Search button
61 Search result display system
62 Impression expression noun phrase adverb phrase set module
63 Control module
64 Difference extraction module
65 Emphasis module
71 Text data
72 set data
73 Control data
74 Difference data
77 Search result screen
78 Text display column
79 Summary column
80 First search result display field
81 Second search result display field
90 Merge Module
91 Impression expression word / noun phrase set extraction module
92 Control Module A
93 Evaluation viewpoint selection module
94 Control Module B
95 Noun Phrase Pickup Module
96 calculation module
97 Most Frequent Noun Phrase Module
98 Flag setting module
101 Impression expression words and noun phrases
102 Notation shaking control dictionary
103 Thesaurus
104 Impression expression words and noun phrases after notation control
110 Flag setting system
111 Flag setting table
121 Flag setting table

Claims (6)

A search result acquisition means for acquiring search data with a search order;
Document data acquisition means for acquiring document data included in search data in the search result acquisition means;
Extraction means for extracting impression expression words and noun phrases linked to the impression expression words for each of the search data from the document data acquired by the document data acquisition means;
Control means for controlling the impression expression word and the noun phrase extracted by the extraction means;
Using the set of impression expression words and noun phrases controlled by the control means, the search rank is determined from the set of impression expression words and noun phrases included in the search data having the search rank of second or lower. Difference means for acquiring the impression expression word and the noun phrase set excluding the impression expression word and the noun phrase set included in the first search data for each search data of the second or lower search order;
Presenting means for presenting the set of the impression expression word and the noun phrase acquired by the difference means with respect to the search data of the second or lower search rank;
A search result presentation apparatus comprising:   2. The search result presentation device according to claim 1, wherein, when there is a degree adverb associated with the impression expression word and the noun phrase, the extraction unit extracts the degree adverb as a set. An emphasis presenting means for emphasizing and presenting at least one of a degree adverb included in the document data, the impression expression word, or a noun phrase associated with the impression expression word;
Summary presentation means for presenting a summary of the document data using at least one of a degree adverb, the impression expression word, or a noun phrase linked to the impression expression word among the document data;
The search result presentation device according to claim 1, further comprising at least one of them.   Retrieval using the degree adverbs, impression expressions, or noun phrases associated with the impression expressions presented by at least one of the presenting means, the emphasis presenting means, or the summary presenting means The search result presentation apparatus according to claim 3, wherein a request sentence can be created. Retrieval result presentation method used when presenting a retrieval result in a retrieval result presentation apparatus comprising retrieval result acquisition means, document data acquisition means, extraction means, control means, difference means, and presentation means Because
A first step in which the search result acquisition means acquires search data assigned a search order;
A second step in which the document data acquisition means acquires the document data included in the search data in the first step;
A third step in which the extraction means extracts a noun phrase linked to the impression expression word and the impression expression word for each search data from the document data acquired in the second step; and
A fourth step in which the control means controls the impression expression word extracted in the third step and the noun phrase;
The difference means uses the impression expression word and the noun phrase controlled in the fourth step, and the set of the impression expression word and the noun phrase included in the search data of which the search rank is second or lower. To obtain the set of the impression expression word and the noun phrase excluding the set of the impression expression word and the noun phrase included in the search data having the first search order for each search data of the second or lower search order. A fifth step to:
A sixth step in which the presenting means presents a set of the impression expression word and the noun phrase acquired in the fifth step with respect to search data of the second or lower search rank;
A search result presentation method characterized by comprising: A search result acquisition function that acquires search data with a search order,
A document data acquisition function for acquiring document data included in search data by the search result acquisition function;
An extraction function for extracting an impression expression word and a noun phrase linked to the impression expression word for each search data from the document data acquired by the document data acquisition function;
A control function for controlling the impression expression word and the noun phrase extracted by the extraction function;
Using the impression expression word and the noun phrase set controlled by the control function, the search order is determined from the set of the impression expression word and the noun phrase included in the search data having the second or lower search order. A difference function for acquiring the impression expression word and the noun phrase set excluding the impression expression word and the noun phrase set included in the first search data for each search data of the second or lower search order;
With respect to the search data of the search rank second or lower, a presentation function for presenting the set of the impression expression word and the noun phrase acquired by the difference function;
Search result presentation program for realizing the above on a computer

Images