JP2006164028A - Window display device and its method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problem that a list of documents corresponding to a category can be displayed when the category is selected, however, when a document is selected as inverse operation, it is impossible to display all categories, to which the document belongs, by tracing from the document side. <P>SOLUTION: The device is provided with a document classification result file indicating the correspondence relation between categories of all category systems and documents to which the categories belong. The classification result file can be searched similarly from the category side or the document side even if either of the category side and the document side is selected. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a way of displaying two related object groups in a window system. In particular, in an automatic document classification apparatus that classifies an input document into a given category, a category group and a document group are provided. The present invention relates to a graphical user interface in which this method is applied to the relationship.

There has been an apparatus capable of classifying and searching documents (for example, Patent Document 1).
Japanese Patent Laid-Open No. 10-027125

  In an automatic document classification system that sorts input documents into categories defined by the user, for example, to display the current classification status for a given set of documents, the display window is divided into two parts, for example, and a category is displayed on the left side. It is assumed that each category of the group is displayed, and one arbitrary category can be selected with a mouse or a keyboard, and all documents classified into the category selected on the left side can be displayed on the right side.

  One document does not necessarily belong to one category because of the multifaceted nature of the document and ambiguity of categorization. The automatic classification system does not narrow down to a single category and distributes it to multiple categories. If there is a possibility, the correspondence between the category group and the document group is a many-to-many relationship. That is, when one category is determined, a plurality of documents belonging to the category are generally determined. Conversely, when one document is determined, a plurality of categories to which the document belongs is generally determined.

  Further, even in a system that does not allow affiliation to a plurality of categories and is designed so that every document always belongs to only one category, if a plurality of category groups are defined, one document is still used. A plurality of categories correspond to.

  In addition, a plurality of category groups means that when a system is shared by a plurality of users and a user wants to execute classification using a different category system for each user, or for the same user. This means that if you want to coexist multiple classification systems from different viewpoints, you can prepare multiple sets of categories in one system and distribute and hold the same document group in multiple ways. Yes.

  In such a situation, in the conventional system, although all documents belonging to a certain category can be displayed, there is no means for simultaneously displaying all categories to which a certain document belongs.

  Even if multiple category systems are defined, you can understand the classification status in the category system you are currently interested in, but you want to see the classification status in a different category system for the document you are currently interested in. The only way to refocus on another category system was, and then the situation was such that the classification status in the category system that was initially focused on could not be seen. That is, when one wants to know what category a document in a certain category is classified in the category system from another viewpoint, it has conventionally been necessary to search among documents belonging to each category of the category system from another viewpoint. It was. This is unavoidable as long as it takes a hierarchical tree display starting from a category rather than a document-centered display method. On the other hand, in the display method starting from individual documents, it is easy to know what category each document is assigned to in a certain category system, but it is inconvenient to know the tendency of the entire category system.

  Also, within a single category system, when a document is classified into multiple categories, it is not possible to select a document from the right window and display all the categories to which it belongs in the left window. It was.

  The present invention is for solving the above problems. In the conventional document classification system, one category in one category system is first selected, and a list of documents belonging to the category is selected. In contrast to this, in the present invention, it is possible to refer to all corresponding categories starting from a document at the same time, regardless of the category system.

  In other words, when a document displayed on the document display means is selected, all categories to which the document belongs are displayed correspondingly in the category display means, so that a plurality of category systems or a document belongs to one category. Even in a system that is not limited, all categories to which a document belongs can be comprehensively and simultaneously known.

  In this way, if you want to display all the categories to which the document belongs across multiple category systems, you can refer to the classification examples of users who are different from you, or comprehensively understand the classification status from different viewpoints. Sometimes effective.

  For example, at the time of initial learning of the system, the category creator may be at a loss as to which category it is appropriate to put a document. At that time, it is possible that other users have already assigned categories to the same document according to different category systems. Knowing the categories classified by other users makes it possible to refer to the classifications in their own category system.

  In order to solve the above problems, in the present invention, there are two object sets in which each object corresponds many-to-many, and an arbitrary object in each object set is divided into two windows or one divided into two. In the system that can display each of the windows, if one window is the window 1 and the other window is the window 2, when the object of the window 1 is selected, all the corresponding objects are displayed in the window 2, and conversely the window 2 When the object is selected, all the objects of the window 2 corresponding to the selected object are displayed.

  In particular, in an automatic document classification system that distributes input documents to categories defined by the user, a category defined by the user is displayed in the window 1 in order to display the current classification status for a given document set. A window-type display device capable of displaying a document classified into a category corresponding to window 2, and selecting a document displayed in window 2 highlights all the categories corresponding to the selected document. It has a characteristic display method.

  Even in the conventional automatic document classification system, when a category is selected, it is possible to display a document list corresponding to the category. However, when a document is selected as the reverse operation, as shown in FIG. Since the classification result file was not searched, it was impossible to display all categories to which the document belongs from the document side.

  However, according to the present invention, a document classification result file showing the correspondence between categories belonging to all category systems and their affiliated documents is provided, and the classification result file is searched in the same manner regardless of which side of the category or document is selected. Therefore, not only can the entire corresponding document be referred to starting from the category, but also the operation in the opposite direction, that is, the entire corresponding category can be viewed starting from the document.

  As explained in the example, if all categories to which the document belongs across multiple category systems are displayed, you can refer to the classification examples of users who are different from yours or combine classification statuses from different viewpoints. Effective when you want to grasp.

  For example, at the time of initial learning of the system, the category creator may be at a loss as to which category it is appropriate to put a document. At that time, it is possible that other users have already assigned categories to the same document according to different category systems. By knowing the categories classified by other users, it is possible to expect effects such as being able to refer to the classification in the category system of one's own user.

Example 1
Embodiments of the present invention will be described with reference to flowcharts and the like.

  FIG. 1 shows the overall configuration of this system.

  FIG. 2 shows in detail the configuration of the display unit in FIG.

  FIG. 3 is a flowchart of the learning process in this system. In the learning process, a learning document whose classification destination is designated in advance by the user is analyzed, and a dictionary necessary for performing classification is created. First, the learning document is subjected to morphological analysis, and word cutting is performed. Then, common nouns, proper nouns, sa-changing nouns, and unknown words that are not in the morpheme dictionary are picked up, and only the words with a high total appearance frequency and high reliability over all the learning documents are selected and the rest are discarded. Of the remaining words, only the words that appear biased in a specific category are used as valid word candidates, so that a locality is calculated for each picked up word. First, among the learning documents belonging to category Cj, the ratio of documents including word Wi, that is, the number of documents belonging to category Cj and including Wi divided by the number of documents belonging to category Cj is set as Pij. However, the normalization is performed so that the value obtained by adding the localization to all categories is 1 for any word. That is, when all categories are C1, C2,..., Cm, Σ (j = 1, m) Pij = 1. The localization E (Wi) of the word Wi with respect to the category Cj is defined by E (Wi) = 1 + Σ (j = 1, m) Pij * logm Pij. N words are selected in descending order of the localization value, and these are set as valid word candidates (process 301 in the flowchart in FIG. 3).

Next, in order to obtain the distance between words, a vector value called a semantic vector is defined for each effective word. All effective words themselves are taken as vector axes, and co-occurrence probabilities between the effective words and effective words as axes are adopted as the component values. However, the co-occurrence probability of the word Wi with respect to the word Wj is defined by the ratio of the number of documents including both Wi and Wj to the number of documents including Wi. FIG. 5 shows co-occurrence probabilities, for example, the vector representation of the word “trading company” is
(1.0 0.2 0.1 0.6 0.8 0.0 0.0 0.0 0.3 ...)
It becomes.

  Based on these pieces of information, a valid word dictionary is created (process 302 in the flowchart of FIG. 3). FIG. 6 shows an example of a valid word dictionary, which uses the word “trade company” as a headline. The number of belongings indicates how many documents including the word belong to each category, and the number of appearances is the total number of learning documents including the word.

  Once the valid word dictionary is created, the vector representation of the document can be defined by calculating the weighted average of the semantic vectors of the valid words contained therein for each learning document. Further, for each category, a representative vector of the category is determined by taking the average of the vectors of all the learning documents belonging to the category (process 303 in FIG. 3 in the flowchart).

  The effective word weight takes into consideration two points: how effective the effective word itself is for the act of classification, and how important the effective word indicates in each document. And define. The first point of view represents the degree of belonging to each category, and is an idea that the effective word having a higher degree of characterizing a specific category is given a higher weight, and the degree of localization described above is used. The second viewpoint evaluates aspects such as how the valid word is used in the target document and how it is related to the content of the document. The evaluation item is created in advance by paying attention to the case role of the effective word and the linguistic role such as the modification type, and the weight value given when the effective word satisfies the condition of each evaluation item is shown in the flowchart of FIG. It is determined by repetitive learning as shown in processes 304 to 310. All weights are initialized to 1 at the start of learning.

  The weight based on the second viewpoint stores the value of each evaluation item as a dictionary, and refers to the weight based on the first viewpoint, that is, the effective word dictionary that describes the localization of the effective word, at the time of classification execution, The total weight of valid words is calculated. FIG. 7 shows an example of a weight dictionary storing weight values with the appearance position of a valid word as an evaluation item, and FIG. 8 shows the weights with the linguistic roles such as the case role and the modification type of the valid word as evaluation items. It is an example of a weight dictionary storing values.

  Next, the weight learning described above will be described. While referring to the valid word dictionary created so far, the category representative vector, the weight dictionary describing the current weight value, etc., classification is performed for each of the learning documents, and the weight value for each evaluation item is determined. The final weight value is determined by repeating the process of fine adjustment and re-classification (processes 304 to 305 in the flowchart of FIG. 3). FIG. 4 is a flowchart for explaining the classification process 305.

  First, as performed at the beginning of the learning process, the classification target document is once again subjected to morphological analysis, and the effective word contained therein is picked up by referring to the effective word dictionary (process 401 in FIG. 4). Note that FIG. 4 describes a process for classifying a document other than the learning document after the weight learning process is completed. Therefore, when performing the classification in the weight learning process as described above, FIG. Needless to say, it is not necessary to perform the morphological analysis again, and the result of the morphological analysis performed at the beginning of the learning process may be stored and used as the processing result here.

  Next, the weight of the picked-up valid word is calculated (process 302 in the flowchart in FIG. 3). The weight value according to the first viewpoint uses the locality of the effective word dictionary, and the weight according to the second viewpoint refers to the two weight dictionaries in FIGS. Find the weight.

  Next, a semantic vector is acquired from the effective word dictionary (process 303 in the flowchart in FIG. 3). When the above is performed for all the effective words picked up from the classification target document, the weight vector calculated in 302 is added to the semantic vector of each effective word, and the average is obtained to obtain the document vector of the classification target document (processing in the flowchart in FIG. 3). 304).

  Then, the distance between the representative vector of each category obtained in the process 303 in FIG. 3 and the document vector of the classification target vector is calculated (process 405 in the flowchart in FIG. 4). As a distance between two vectors, a cosine value of an angle formed by both vectors obtained by calculating an inner product is used as is often done. The category having the shortest distance is determined as the classification target category of the classification target document (process 406 in the flowchart in FIG. 4).

  The above is the classification execution process performed at 305 in the flowchart of FIG. 3, and the continuation of the weight learning will be described by returning to the process 306 of FIG.

  Since the correct category that the user wants to classify is determined in advance in the learning document, it is compared with the classification destination category obtained in the processing 305. If the two categories match, it is considered that the weight need not be tuned, and the learning document is not processed any further, and the process proceeds to the processing of the learning document next to the loop 304. If the two categories do not match, it means that the classification destination of the classification system is wrong, and the weight value of the weight dictionary is adjusted as follows. First, with respect to the valid words included in the learning document, the semantic vector acquired in 403 of the classification process is referred to, and the distance between the correct answer category and the classification category is calculated. It is determined which of the category and the classification destination category is closer to the representative vector (process 308 in FIG. 3 in the flowchart). If it is close to the classification destination category, the document vector of the document is closer to the correct category, so that the two weight dictionaries reduce the weight value of the evaluation item of the weight corresponding to the valid word by a minute amount. Is corrected (processing 309 in the flowchart of FIG. 3). Conversely, if the document vector of the document is closer to the correct category, the weight evaluation item to which the valid word corresponds is considered to have contributed to the correct classification, and the weight value is increased by a small amount. Thus, the weight dictionary is corrected (process 310 in the flowchart of FIG. 3). This is performed for all valid words picked up from the learning document (loop 307 in the flowchart of FIG. 3), and the weight value is adjusted.

  Then, the above-described processing is repeated with all learning documents regarded as classification targets (loop 304 in the flowchart in FIG. 3), and several times for a series of learning documents so that the weight values are optimum. Repeat to finally complete the weight dictionary. In this way, the learning process is completed.

  The general documents other than the learning document are classified by referring to the various dictionaries created as described above, and the procedure is exactly the same as the classification execution process described in the weight learning process. The only difference is whether the classification target document is a learning document to which a correct category is given or whether the correct category is an unknown general document.

  However, in the process 406 of the flowchart in FIG. 4, only one classification destination category of the classification target document is determined and output. However, the distance between the category representative vector and the document vector of the classification target document happens to coincide. There may be a plurality of categories that can be used. In addition, depending on the system, it may be required to output a plurality of classification destination categories. In response to the latter request, a correct answer may be used if the distance between the category representative vector and the document vector of the classification target document exceeds a predetermined threshold.

  As a result of classifying general documents in this way, the user determines whether the category determined by the system is correct or incorrect, and if necessary, designates the correct category considered by the user and causes the system to learn. Therefore, it is required to display the classification result in an easy-to-understand manner, and the display method proposed in the present invention will be described next.

  FIG. 9 is an example in which the classification categories defined on the automatic document classification system by the system administrator or the end user of the system are displayed on the left side of the display window in the present invention. FIG. 10 shows a category definition file, which is an example of a system file for storing the contents of categories and category systems defined by a system administrator for automatic document classification. Each entry consists of a set of three values, each representing a category system ID, the number of categories defined in the category system, and a list of category IDs defined in the category system.

  A single classification system may be used by multiple end users, and you may want to classify a group of documents in different category systems for each user. Since they exist, here, the category systems are distinguished by the names of category set A, category set B, and the like. In the category system of category set A, there are eight categories, “politics”, “economy”, “judicial”, “education”, “medical”, “literary”, “academic”, and “incidents”. In addition, the category set B includes six categories of “human”, “thing”, “entertainment”, “education”, “current affairs”, and “others”. Show.

  FIG. 11 shows an example of a classification result file, which is a file describing which category of which category system each document is classified with respect to a group of documents for which automatic classification has been executed by this system. Updated each time classification is performed on the target document. Each entry includes a “document ID attribute” for uniquely identifying a document stored in the system, a “category system attribute” indicating an ID of a category system of interest, and an ID of a category to which the document belongs in the category system. It is expressed as a set having three attributes “category attribute”.

  In FIG. 9, when an arbitrary category is selected by a mouse, a keyboard, or the like, the display control device searches the entry having the category attribute equal to the selected category with reference to the classification result file of FIG. To display the document title.

  FIG. 12 shows an example of the screen at that time, and is a display example when the “politics” category in the category set A is selected from the category list shown in FIG. The right side of the window is for displaying a list of documents classified in the category selected on the left side of the window. In this example, the titles of all documents belonging to the “politics” category are displayed.

  Next, when an arbitrary document is selected from the list of document titles displayed in the right window of FIG. 12 with the mouse or the keyboard, the display control apparatus refers again to the classification result file of FIG. 11 and the document ID attribute is selected. All entries that are equal to the document are searched, and a pair of category system attribute and category attribute value of the entry is acquired. Then, only the category corresponding to the acquired pair is highlighted from the category list displayed in the right window.

  Even in the conventional automatic document classification system, when a category is selected, it is possible to display a document list corresponding to the category. However, when a document is selected as the reverse operation, as shown in FIG. Since the classification result file was not searched, it was impossible to display all categories to which the document belongs from the document side. In the present invention, a document classification result file indicating the correspondence relationship between the categories over all category systems and their belonging documents is provided, and the classification result file is similarly searched regardless of which side of the category or document is selected. It is possible not only to refer to all corresponding documents starting from a category, but also to perform an operation in the opposite direction, that is, to view all corresponding categories starting from a document.

  FIG. 13 shows an example of the screen at this time. When a document whose title is “Whereabouts of Structural Reform” is selected from the document list, the category in which this document is classified is highlighted in the left window. Indicates that it is lighted. In the category set A, the document is classified into two categories of “politics” and “economy”, and in the category set B, it is classified into one “current affairs” category. The category is highlighted. However, unless the “politics” category of category set A, which is the category of interest, is displayed separately from the other two categories, the document group listed in the right window belongs to where it belongs. For example, the “politics” category can be distinguished from the corresponding category by changing the display color or displaying the underline.

  Even if the above operation method is adopted, it is compatible with the document content display function that has been conventionally realized. That is, in order to display the full text of the selected document, the viewer can be activated by double-clicking the mouse on the document or clicking the right button of the mouse. 14 to 15 show the contents of the document displayed by this operation.

(Example 2)
In the first embodiment, when a category is selected, a document list corresponding to the category is displayed, or when a document is selected as the reverse operation, all categories to which the category belongs are displayed. A single classification result file was searched in common. In FIG. 11, since each entry of the classification result file uses the attribute of the document ID as a headline, searching the corresponding category when the document is selected can be processed at high speed, but conversely when the category is selected. Pulling all the corresponding documents is a relatively time-consuming process.

  Therefore, another classification result file as shown in FIG. 16 is created and updated together with the classification result file of FIG. 11 every time classification is performed on a certain classification target document. FIG. 16 is a table in which the headings of the entries are rearranged so as to have the category attributes with the same contents as FIG.

(Other examples)
It should be noted that the present invention as described above can take the form of, for example, a system, apparatus, method, program, or storage medium, and is specifically applied to a system composed of a plurality of devices. Alternatively, it may be applied to an apparatus composed of one device.

  In the present invention, a software program (in the embodiment, a program corresponding to the flowchart shown in the figure) that realizes the functions of the above-described embodiment is directly or remotely supplied to the system or apparatus, and the computer of the system or apparatus Is also achieved by reading and executing the supplied program code.

  Accordingly, since the functions of the present invention are implemented by computer, the program code installed in the computer also implements the present invention. In other words, the present invention includes a computer program itself for realizing the functional processing of the present invention.

  In that case, as long as it has the function of a program, it may be in the form of object code, a program executed by an interpreter, script data supplied to the OS, or the like.

  As a recording medium for supplying the program, for example, floppy (registered trademark) disk, hard disk, optical disk, magneto-optical disk, MO, CD-ROM, CD-R, CD-RW, magnetic tape, nonvolatile memory card ROM, DVD (DVD-ROM, DVD-R) and the like.

  As another program supply method, a computer browser is used to connect to a homepage on the Internet, and the computer program itself of the present invention or a compressed file including an automatic installation function is downloaded from the homepage to a recording medium such as a hard disk. Can also be supplied. It can also be realized by dividing the program code constituting the program of the present invention into a plurality of files and downloading each file from a different homepage. That is, a server that allows a plurality of users to download a program file for realizing the functional processing of the present invention on a computer is also included in the present invention.

  In addition, the program of the present invention is encrypted, stored in a storage medium such as a CD-ROM, distributed to users, and key information for decryption is downloaded from a homepage via the Internet to users who have cleared predetermined conditions. It is also possible to execute the encrypted program by using the key information and install the program on a computer.

  In addition to the functions of the above-described embodiments being realized by the computer executing the read program, the OS running on the computer based on an instruction of the program is a part of the actual processing. Alternatively, the functions of the above-described embodiment can be realized by performing all of them and performing the processing.

  Furthermore, after the program read from the recording medium is written in a memory provided in a function expansion board inserted into the computer or a function expansion unit connected to the computer, the function expansion board or The CPU or the like provided in the function expansion unit performs part or all of the actual processing, and the functions of the above-described embodiments are realized by the processing.

  Even in the conventional automatic document classification system, when a category is selected, it is possible to display a document list corresponding to the category. However, when a document is selected as the reverse operation, as shown in FIG. Since the classification result file was not searched, it was impossible to display all categories to which the document belongs from the document side.

  However, according to the present invention, a document classification result file showing the correspondence between categories belonging to all category systems and their affiliated documents is provided, and the classification result file is searched in the same manner regardless of which side of the category or document is selected. Therefore, not only can the entire corresponding document be referred to starting from the category, but also the operation in the opposite direction, that is, the entire corresponding category can be viewed starting from the document.

  As explained in the example, if all categories to which the document belongs across multiple category systems are displayed, you can refer to the classification examples of users who are different from yours or combine classification statuses from different viewpoints. Effective when you want to grasp.

  For example, at the time of initial learning of the system, the category creator may be at a loss as to which category it is appropriate to put a document. At that time, it is possible that other users have already assigned categories to the same document according to different category systems. By knowing the categories classified by other users, it is possible to expect effects such as being able to refer to the classification in the category system of one's own user.

It is the figure which showed the whole structure of the system in this invention. It is the figure which showed the structure of the display part in this invention. It is a flowchart showing the flow of the learning process in this invention. It is a flowchart showing the flow of the classification process in this invention. It is the figure which showed the example of the co-occurrence probability between words. It is the figure which showed the example of the effective word dictionary. It is an example of the weight dictionary which used the positional role of the effective word as an evaluation item. It is an example of the weight dictionary which used the linguistic role of the effective word as an evaluation item. It is the example which displayed the category list on the display window in this invention. It is the figure which showed the example of the category definition file. It is the figure which showed the example of the classification result file which uses document ID as a headline. It is the figure which showed the example of a screen when a category is selected in a display window. It is the figure which showed the example of a screen when a document is selected in a display window. It is the figure which showed the example of a screen when a document is right-clicked in a display window. It is the figure which showed the example of a screen when the text of a document is displayed on the display window. It is the figure which showed the example of the classification result file changed into the category headline used in another Example.

Claims (8)

  There are two object sets in which each object corresponds to many-to-many, and in a system that can display an arbitrary number of objects in each object set in two windows or one divided window, When the window 1 is the window 2 and the other window is the window 2, when the object of the window 1 is selected, all the corresponding objects are displayed in the window 2. Conversely, when the object of the window 2 is selected, all the corresponding windows 2 are displayed. A window-type display device characterized in that the object is displayed.   2. The window type display device according to claim 1, wherein a current classification status for a given document set is displayed in an automatic document classification system that distributes an input document to a category group defined by a user. An automatic document classification apparatus that displays a document defined by a user in a category corresponding to window 2, and when a document displayed in window 2 is selected, all categories to which the document belongs are highlighted. An automatic document classification device characterized by being displayed.   3. The automatic document classification apparatus according to claim 2, wherein when there are a plurality of categories to which a certain document belongs, a case where the classification system represents that the document is classified into a plurality of categories within the same category system, and An automatic document classification device characterized in that a display method is changed depending on whether it belongs to a plurality of different categories.   3. The automatic document classification apparatus according to claim 2, wherein when there are a plurality of categories to which the document belongs, the display form, for example, the display color of the category is changed in the order of the category most suitable as a document classification destination. Automatic document classification device.   3. The automatic document classification apparatus according to claim 2, wherein a correspondence category between a document displayed in window 2 and a category displayed in window 1 is displayed in a category system corresponding to a selected document, and a plurality of categories are displayed. An automatic document classification device characterized in that it can switch between simultaneous display methods of categories across category systems.   There are two object sets in which each object corresponds to many-to-many, and in a system that can display an arbitrary number of objects in each object set in two windows or one divided window, When the window 1 is the window 2 and the other window is the window 2, when the object of the window 1 is selected, all the corresponding objects are displayed in the window 2. Conversely, when the object of the window 2 is selected, all the corresponding windows 2 are displayed. A window type display method characterized in that an object is displayed.   There are two object sets in which each object corresponds to many-to-many, and in a system that can display an arbitrary number of objects in each object set in two windows or one divided window, When the window 1 is the window 2 and the other window is the window 2, when the object of the window 1 is selected, all the corresponding objects are displayed in the window 2. Conversely, when the object of the window 2 is selected, all the corresponding windows 2 are displayed. A computer program for causing a computer to execute an object so that the object is displayed.   A computer-readable storage medium storing the computer program according to claim 7.

Images