JP2006092091A - Document structuring device and document structuring method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a document structuring device allowing reduction of cost spent for production of an extraction rule or a document structure-extracting logic model prepared in each target document, and allowing improvement of structure extraction accuracy by only performing browsing operation of the document. <P>SOLUTION: This document structuring device has: a structure recognition means 102 recognizing a document structure from stored document data 101a, and storing it as document structure information 103a; a distribution means 104 distributing the stored document data to a client terminal 105, acquiring a history of a browsing part of a user, and storing it as browsing history information 106a; and a correlation calculation means 107 calculating correlation between document structure blocks from the document structure information and the browsing history information 106a. The structure recognition means 102 receives the correlation between the document structure blocks calculated by the correlation calculation means 107, and re-recognizes the document structure. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an apparatus and method for recognizing a document structure from document image data and converting it into structured text, and more particularly to a document structuring apparatus and document for improving a document structure recognition result by using a browsing history of a document viewer. It relates to a structuring method.

  When paper documents are digitized for distribution and reuse, character recognition and document structure (semantic structures that make up documents such as titles, chapters, sections, paragraphs, etc.) are recognized, and searches using structural information It is considered to perform (for example, a document including “Internet” in the title) and partial distribution (for example, distribution of only the second paragraph of the document). By using the structure information, the former can improve the search accuracy, and the latter can improve the communication efficiency. However, the degree of improvement largely depends on the recognition result of the document structure.

  Various methods have been proposed for document structure recognition from document data. For example, in Patent Document 1, document structure recognition is performed using document layout information. In Patent Document 2 and Patent Document 3, document structure recognition is performed using a pattern match of a character string recognized as an extraction rule. Patent Document 4 proposes a method of automatically adding structure information using an operation history (copy, paste, etc.) when a document is created by a word processor or the like.

On the other hand, in Patent Document 5, in order to generate a term-related network that sufficiently reflects the intention of the document user, the document displayed based on the display history of the document displayed on the document display device by the document user. Analyzes the relationship between the display transitions for each part, calculates the relevance between terms extracted from the document part having the display transition relationship based on the display transition relationship, and associates terms with a term relevance greater than or equal to a predetermined value A method for generating a terminology related network is described. However, this Patent Document 5 does not describe that the display history is particularly used for the analysis of the document structure itself.
Japanese Patent Laid-Open No. 11-184894 JP-A-8-287189 JP-A-9-319747 Japanese Patent Laid-Open No. 10-222506 Japanese Patent Laid-Open No. 10-222532

  In the document structuring method of Patent Document 4, since an operation history at the time of document creation is used, the system required by Patent Document 4 must be used from the document creation stage, and has already been completed or output to paper. There is a problem that it cannot be applied to a document that has been processed.

  On the other hand, the document structuring methods of Patent Documents 1, 2, and 3 that extract a document structure using a logical model of a document layout and a structure extraction rule do not cause such a problem. However, there is a problem in that the recognition accuracy is low if the logical model of the document layout and the structure extraction rule that are the norm for recognition do not match the target document accurately. In order to improve the recognition accuracy, conventionally, it has been necessary to accurately match the logical model of the document layout and the structure extraction rule as the reference document with the target document, which requires a lot of labor. In addition, the document layout logical model and structure extraction rules that are precisely matched to a specific target document are specific to the target document. There is a problem that it is lowered and lacks versatility.

  SUMMARY OF THE INVENTION An object of the present invention is to provide a novel document structuring apparatus and document structuring method capable of improving the recognition accuracy of a document structure without depending on a normative document layout logical model or structure extraction rule.

  The document structuring apparatus according to the present invention includes a document storage unit that stores document data, a document structure storage unit that stores document structure information of the document data, a browsing history storage unit that stores browsing history information, and the document storage. When the document data stored in the means is distributed to the document browsing terminal, the history of the browsing position of the document data is acquired and stored as browsing history information in the browsing history storage means, and the stored browsing history information is used. And calculating a correlation between the document structure blocks in the document structure information of the document data, and re-recognizing the document structure of the document data using the calculated correlation result between the document structure blocks. It is characterized by. More specifically, the first document structuring apparatus of the present invention includes document storage means for storing document data, document structure storage means for storing document structure information, and document data stored in the document storage means. A structure recognizing means for recognizing the document structure and storing the document structure information in the document structure storage means, a browsing history storage means for storing browsing history information, and referring to the stored document structure information A distribution unit that distributes a part or all of the document to the document browsing terminal; a browsing history acquisition unit that acquires a history of a browsing portion of the document distributed to the document browsing terminal and stores the history as browsing history information in the browsing history storage unit; Correlation calculation means for calculating correlation between document structure blocks in the stored document structure information using the stored browsing history information, the structure recognition means, Serial and wherein the correlation calculation means performs a re-recognition of the received document structure correlations between document structure blocks calculated.

  According to a second document structuring apparatus of the present invention, in the first document structuring apparatus, the structure recognition unit updates the stored document structure information based on a result of re-recognition. And

  According to a third document structuring apparatus of the present invention, in the first or second document structuring apparatus, the distribution unit distributes document data having a size corresponding to a display area of the document browsing terminal in one distribution. The browsing history acquisition means acquires the information of the document structure block included in the distributed document data as a history of browsing locations of the document.

  According to a fourth document structuring apparatus of the present invention, in the third document structuring apparatus, the browsing history acquisition unit includes time information at which document data is distributed in browsing history information. .

  According to a fifth document structuring apparatus of the present invention, in the first to fourth document structuring apparatuses, the correlation calculating unit is configured to obtain a certain number of the browsing history information, every time a certain time elapses, or The correlation between the document structure blocks is calculated whenever there is an instruction from the document browsing terminal, or at an arbitrary combination timing.

  According to a sixth document structuring apparatus of the present invention, in the first or second document structuring apparatus, the correlation calculating unit includes a plurality of browsing history information relating to the same document stored in the browsing history storing unit. In addition, when there are a plurality of browsing location transitions starting from the same document structure block, the correlation between the document structure blocks is calculated based on the transition of the browsing location occupying a larger number.

  In the first document structuring method of the present invention, a) the structure recognizing means recognizes the document structure from the document data stored in the document storage means, and accumulates it in the document structure storage means as document structure information; b) A step of distributing part or all of the document data to the document browsing terminal by referring to the stored document structure information; c) a document distributed by the browsing history acquiring unit to the document browsing terminal D) a step of acquiring a history of browsing locations and storing the history as browsing history information in browsing history storage means; d) a correlation calculation means using the stored browsing history information to store the document structure in the stored document structure information A step of calculating a correlation between blocks; e) a step in which the structure recognition unit receives the correlation between the document structure blocks calculated by the correlation calculation unit and re-recognizes the document structure. , Characterized in that it comprises a.

  The second document structuring method of the present invention includes the step of f) the structure recognizing unit updating the accumulated document structure information based on a result of re-recognition in the first document structuring method. It is characterized by that.

  According to a third document structuring method of the present invention, in the first or second document structuring method, the distribution unit distributes document data having a size corresponding to a display area of the document browsing terminal in one distribution. The browsing history acquisition means acquires the information of the document structure block included in the distributed document data as a history of browsing locations of the document.

  According to a fourth document structuring method of the present invention, in the third document structuring method, the browsing history acquisition unit includes time information at which the document data is distributed in the browsing history information.

  According to a fifth document structuring method of the present invention, in the first to fourth document structuring methods, the correlation calculating means is configured to obtain a certain number of the browsing history information, or to elapse a certain time, or The correlation between the document structure blocks is calculated every time there is an instruction from the document browsing terminal, or at an arbitrary combination timing.

  According to a sixth document structuring method of the present invention, in the first or second document structuring method, the correlation calculating means includes a plurality of browsing history information relating to the same document stored in the browsing history storage means. In addition, when there are a plurality of browsing location transitions starting from the same document structure block, the correlation between the document structure blocks is calculated based on the transition of the browsing location occupying a larger number.

"Action"
In the present invention, when the document data is distributed to the document browsing terminal, the history of the browsing portion of the document data is acquired and stored as browsing history information, and the document is stored using the stored browsing history information. The correlation between the document structure blocks in the document structure information of the data is calculated, and the document structure of the document data is re-recognized using the calculated correlation result between the document structure blocks. More specifically, the structure recognizing unit recognizes the document structure such as a document structure block which is a group of areas such as titles and paragraphs from the document data stored in the document storage unit, and the document structure information is obtained as document structure information. Accumulating in the storage means, the distribution means refers to the stored document structure information and distributes part or all of the document data to the document browsing terminal, and the browsing history storage means distributes to the document browsing terminal The browsing history of the document is acquired and stored as browsing history information in the browsing history storage means, and the correlation calculating means uses the stored browsing history information to store the document structure blocks in the stored document structure information. The correlation is calculated, and the structure recognizing means receives the correlation between the document structure blocks calculated by the correlation calculating means and re-recognizes the document structure. The user of the document browsing terminal tends to continue reading the document structure block, and therefore, the continuity between document structure blocks, which is a kind of document structure, is detected from the history of the document browsing location. It is possible to improve the document structure recognition result using this detection result. Thus, according to the present invention, the browsing history information is used to improve the document structure recognition result, so that it is possible to reduce the trouble of creating the logical model and the extraction rule used by the structure recognition unit for each target document. Therefore, the present invention can be applied to a document that has already been completed or output on paper.

  The first effect is that the recognition accuracy of the document structure can be improved without changing the logical model or structure extraction rule of the standard document layout, and the trouble of creating the logical model and the extraction rule for each target document is reduced. It can be reduced. The reason is that the document structure recognition result is improved by using the browsing history information of the document.

  The second effect is versatility. The reason is that the dependence on the target document is low and the document structure recognition result is improved by using the browsing history information of the document.

  The third effect is that it can be applied to a document that has already been completed or output on paper. This is because an operation history at the time of document creation is not required.

  Next, the best mode for carrying out the present invention will be described in detail with reference to the drawings.

“First Embodiment”
Referring to FIG. 1, a document structuring apparatus 110 according to the first exemplary embodiment of the present invention includes a document storage unit 101 that stores document data 101a, a document structure storage unit 103 that stores document structure information 103a, The browsing history storage unit 106 stores the browsing history information 106a, and the processing unit 109 connected thereto. The processing unit 109 can communicate with the client terminal 105 that is one or more document browsing terminals via a network such as the Internet.

  The client terminal 105 includes a mobile phone, a portable information terminal, a personal computer, and the like, and has a function of displaying document data distributed from the distribution unit 104 on a display device and presenting it to the user. Further, it has a function of requesting the distribution means 104 to distribute the next document data in response to a user operation.

  The processing unit 109 includes a workstation, a personal computer, and the like, and recognizes the document structure from the document data 101a stored in the document storage unit 101, and stores the structure in the document structure storage unit 103 as document structure information 103a. A distribution unit 104 that distributes part or all of the document data 101a stored in the document storage unit 101 to the client terminal 105 with reference to the document structure information 103a stored in the document structure storage unit 103, and a client terminal A browsing history acquisition unit 108 that acquires the browsing history of the document data 101 a distributed to 105 and stores the browsing history information 106 a in the browsing history storage unit 106, and the browsing history information 106 a stored in the browsing history storage unit 106. Document structure information 1 stored in the document structure storage unit 103 And a correlation calculation unit 107 for calculating the correlation between the document structure blocks in 3a. The structure recognizing unit 102 receives the correlation between the document structure blocks calculated by the correlation calculating unit 107, and re-recognizes the document structure.

  Next, an outline of the overall operation of the present embodiment will be described with reference to the flowcharts of FIGS.

  The document structuring apparatus 110 first reads out document data 101a obtained by digitizing a document page for recognizing the document structure from the document storage unit 101, and the structure recognizing unit 102 uses a known method in Patent Documents 1 to 3, for example. The document data 101a is divided into several areas (document structure blocks), and the document structure (title, paragraph, etc.) of each area is recognized (step 201 in FIG. 2). The document structure information recognized by the structure recognition unit 102 is stored in the document structure storage unit 103.

  Next, the document structuring apparatus 110 refers to the recognized document structure information from the document structure storage unit 103 by the distribution unit 104, and the entire or a part of the document data 101a stored in the document storage unit 101 is transferred to the client terminal. (Step 202).

  Next, the document structuring apparatus 110 uses the browsing history acquisition unit 108 to acquire the history of browsing locations by the user of the client terminal 105 for the distributed document data (step 203). The acquired browsing history is stored in the browsing history storage unit 106 as browsing history information 106a.

  Next, the document structuring apparatus 110 calculates the correlation between the document structure blocks recognized by the structure recognition unit 102 based on the browsing history stored in the browsing history storage unit 106 by the correlation calculation unit 107 at a predetermined timing. Then, the calculation result is notified to the structure recognition means 102 (step 204). As the above-mentioned predetermined timing, every time browsing history information (n is a predetermined integer equal to or greater than 1) is obtained, every t seconds (t is a predetermined positive numerical value), or Any combination of the timings can be used.

  Next, the document structuring apparatus 110 re-recognizes the document structure of the document data 101 a using the correlation calculated by the correlation calculation unit 107 by the structure recognition unit 102 and updates the document structure information 103 a in the document structure storage unit 103. (Step 205).

  Similarly, steps 202 to 205 may be repeated any number of times using the re-recognized document structure information.

  Next, the effect of this embodiment will be described.

  In this embodiment, since the document structure recognition result is re-recognized using the browsing history information of the document, the document structure is not dependent on the structure recognition logical model used by the structure recognition unit 102 or the modification of the extraction rule. Recognition accuracy can be improved, and the effort of creating a logical model or extraction tool for each target document can be reduced.

  Further, in this embodiment, since an operation history at the time of document creation is not required, document structure recognition can be performed on a document that has been completed or has been output on paper.

  Furthermore, in this embodiment, since the dependence on the target document is low and the document structure recognition result is improved by using the browsing history information of the document, it can be applied to various documents and is versatile.

  Next, examples of the first embodiment of the present invention will be described in detail with reference to the drawings.

  As in the document structuring apparatus 110 according to the first embodiment shown in FIG. 1, the document structuring apparatus according to the present example includes a document storage unit 101 that stores document data 101a and document structure information 103a. A document structure storage unit 103 that stores information, a browsing history storage unit 106 that stores browsing history information 106a, and a processing unit 109 that is connected to these and connected to one or more client terminals 105 through a network such as the Internet. Then, the processing unit 109 recognizes the document structure from the document data 101 a stored in the document storage unit 101 and stores the document structure information 103 a in the document structure storage unit 103 as document structure information 103 a and the document structure storage unit 103. A part of the document data 101a stored in the document storage unit 101 with reference to the stored document structure information 103a Alternatively, a distribution unit 104 that distributes all to the client terminal 105, and a browsing history acquisition unit that acquires the history of browsing portions of the document data 101a distributed to the client terminal 105 and accumulates the browsing history information 106a in the browsing history storage unit 106. 108 and correlation calculation means 107 for calculating the correlation between the document structure blocks in the document structure information 103a stored in the document structure storage unit 103 using the browsing history information 106a stored in the browsing history storage unit 106. ing.

  The document storage unit 101 stores document data 101a targeted by the present embodiment. For example, the document storage unit 101 stores each page of a paper document converted into image data using an input unit such as a scanner.

  The structure recognizing unit 102 receives the image data of each page accumulated as document data as an input, and performs area division and character recognition of the image data using a known unit as shown in Patent Document 2, for example (see FIG. 3), a score for determining the document structure of each area is calculated using the rules shown in FIG. 4 for each divided area (step 302), and finally for each area. The document structure having the maximum score value is determined as the document structure of the area (step 303). The calculated score and the determined document structure are accumulated in the document structure information 103. FIG. 5 shows an example of processing steps of the structure recognition unit 102. For reference, FIG. 6 shows an example in which the document structure, which is the final result of FIG. 5, is expressed by a structured document in XML (eXtensible Markup Language) format.

  Referring to FIG. 5, the structure recognizing unit 102 performs region division and character recognition on the page image to recognize blocks 501 to 505, and the document structure of each block using the rules of FIG. 4 for each block. Is calculated and recorded in the score table ST constituting a part of the document structure information 103a of the document structure storage unit 103. The score table ST has a line for each recognized block, and holds information regarding the block and a score regarding the title, paragraph, and continuation in each line. In this example, since the block 501 is a character having a character size of 22 points, 20 points are given to the title score by the rule R1. Further, since the blocks 502 to 505 have two or more lines, 10 points are given to the score of the paragraph by the rule R2. Furthermore, a continuation score is given to the blocks 503 and 505 by the rules R3 and R4. The document structure table BT also forms part of the document structure information 103a and records the document structure determined for each block.

  In the above description, the determined document structure is stored in the document structure information 103a. However, not only the determined document structure but also region (block) coordinate information, character size information, and character recognition result. Etc. may be accumulated together. Alternatively, it may be stored as a structured document in XML format as shown in FIG.

  The distribution unit 104 refers to the document structure information 103 a and distributes all or part of the image data related to the document data 101 a accumulated in the document storage unit 101 to the client terminal 105. For example, when the client terminal 105 can display an area of the size indicated by the broken line in FIG. 7 for document data having document structure information as shown in FIG. Of the block 501 and the block 502, the image data of the portion overlapping the display area of the client terminal 105, and the block 503 belonging to the same document structure as the block 502, The portion of the image data that overlaps the display area of the client terminal is distributed to the client terminal 105. Further, when the user of the client terminal 105 requests distribution of the image data of the adjacent part by scrolling the screen or the like, the distribution unit 104 cuts out the image data of the adjacent part from the document data 101 a and distributes it to the client terminal 105. .

  On the other hand, each time the distribution unit 104 distributes image data to the client terminal 105, the browsing history acquisition unit 108 acquires area information located at the barycentric coordinates of the distributed area as browsing history information 106a, and the browsing history storage unit 106 To accumulate. FIG. 8 shows an example of the transition of the center-of-gravity position coordinates of the distribution area and the browsing history information 106 a stored in the browsing history storage unit 106.

  The correlation calculation unit 107 calculates the correlation between the blocks recognized by the structure recognition unit 102 using the browsing history information 106a accumulated in the browsing history storage unit 106. For example, the correlation between the blocks listed as the browsing history (in the case of browsing history information shown in FIG. 8, 502 and 505) is +20, and the block on the browsing history and the block not in the browsing history (shown in FIG. 8) In the case of browsing history information, the correlation between 502 and 503, 504 and 505, etc.) is calculated as -20. The correlation between blocks appearing adjacent to the browsing history (in the case of browsing history information shown in FIG. 8, 502 and 505) may be calculated as +20.

  The structure recognition unit 102 re-recognizes the document structure using the correlation calculated by the correlation calculation unit 107. For example, the structure recognizing unit 102 adds the correlation value calculated by the correlation calculating unit 107 to the “continuation” score on the score table ST, and re-recognizes the document structure. For example, when the correlation calculation unit 107 calculates that the correlation between the blocks 502 and 505 is +20, and the correlation between the blocks 502 and 503 and 504 and 505 is −20, the result is “continuation” on the score table ST in FIG. When added to the score, the contents of the score table ST are as shown in FIG. Therefore, when the document structure is re-recognized from this content, it is as shown in the document structure table BT of FIG. 9, and regarding the continuity between blocks, the block 505 is re-recognized as continuing from the block 502. FIG. 10 shows an example in which the final result of FIG. 9 in which the document structure is re-recognized with respect to the processing process example of FIG. 5 is expressed by the structured document in the XML format as in FIG.

  Similarly, the above processing corresponding to steps 202 to 205 in FIG. 2 may be repeated any number of times by using the newly obtained browsing history and the re-recognized document structure information.

  Next, the effect of the present embodiment will be described.

  In this embodiment, since the document structure recognition result is re-recognized using the browsing history information of the document, it is possible to reduce the trouble of creating the structure recognition logical model and the extraction rule used by the structure recognition unit 102 for each target document. it can. For example, among the score calculation rules as shown in FIG. 4, according to this embodiment, correction of the “continuation” score calculation using the browsing history of the document without modifying the rules R3 and R4. Can be performed.

  Further, in this embodiment, since an operation history at the time of document creation is not required, document structure recognition can be performed even for a document that has been completed or output on paper.

  In this embodiment, the distribution unit 104 refers to the document structure information 103a and transmits the image data of the area belonging to the same document structure together. Therefore, the user of the client terminal 105 may need next. Image data can be distributed first. Thereby, there is also an effect that the response of the client terminal 105 can be improved. Further, when the document structure is updated as shown in FIGS. 6 to 10 using the browsing history of the client terminal 105, the image data distributed together when the block 501 and the block 502 are distributed becomes the block The block is changed from 503 to block 505. That is, the image data that the distribution unit 104 distributes to the client terminal 105 is changed to image data that is more likely to be required next in the client terminal 105 according to the browsing history of the client terminal 105. Can be improved.

  In the description of the above embodiment, the document data 101 is obtained by converting each page of a paper document into image data using an input unit such as a scanner. For example, Acrobat is a document editing tool manufactured by Adobe. As described above, image data generated using software capable of generating a page image from electronic document data may be stored. Further, the structure recognition means 102 is data other than image data, for example, a data conversion tool plain2 that can be obtained from “http://ftp.debian.or.jp/debian-jp/pool/main/p/plain2/”. If text data is input to and converted into a document embedded with LaTeX or troff commands, it may be stored in text data format. In these cases, there is an effect that not only a paper document but also electronic document data and text data can be processed.

  In the description of the above embodiment, the structure recognizing unit 102 determines the document structure after performing region division and character recognition on the image data. For example, a known unit as disclosed in Patent Document 1 is used. It is also possible to determine the document structure only by segmentation without using character recognition.

  In the description of the above embodiment, the browsing history acquisition unit 108 accumulates the area information located at the center of gravity of the distributed area as the browsing history information 106a. However, the browsing history information 106a is not the center of gravity coordinates but the area information included in the distributed area. The browsing history may be extracted and stored using the area. For example, as shown in FIG. 11, when the area distributed to the client is C11, the areas C21, C22, and C23 of the blocks C01, C02, and C03 included in the area are compared, and the area C22 is the largest. Block C02 is accumulated as a browsing history. In the case of the distribution area C12, if the blocks C02 and C05 have the same maximum area, both blocks are accumulated as a browsing history. Similarly, in the case of the distribution area C13, the block C05 is accumulated, and finally browsing history information as shown in FIG. 11 is obtained. If you do this, you can extract information about blocks that have a high percentage of the distribution area, so that you can calculate correlations based on information about blocks that users are most likely to browse There is an effect of becoming.

  As the browsing history information, not only the area information determined by the position of the center of gravity and the maximum area described above, but also the time of distribution may be stored at the same time. In this case, based on the delivered time difference, a positive correlation value is given between blocks with a time difference of a certain value or less, and a negative correlation value is given between blocks with a time difference of a certain value or more. You may make it calculate the correlation between blocks. FIG. 12 shows an example of the calculation rule.

  In the description of the above embodiment, the distribution unit 104 distributes only the image data corresponding to the area to be displayed on the client terminal 105. However, the image data may be distributed for the entire document or for each page. In this way, the browsing history acquisition unit 108 cannot obtain the history of the browsing location by the above-described method. In this case, the client terminal 105 stores the operation history such as the user's page movement or screen scrolling. It may be acquired and transmitted as browsing history to the browsing history acquisition unit 108, and the browsing history acquisition unit 108 may store the browsing history information 106a. Also, the movement of the user's line of sight is acquired using a known technique for tracking the user's line of sight to the client terminal 105, and the acquired tracking information of the line of sight is transmitted as a browsing history to the browsing history acquisition means 108, The browsing history acquisition unit 108 may store the browsing history information 106a.

  In the description of the above embodiment, the correlation calculation unit 107 calculates the correlation based on the browsing history information 106a accumulated in the browsing history storage unit 106 by the browsing history acquisition unit 108. When a plurality of different browsing histories are accumulated, the correlation is not calculated by using each history one by one, but is calculated after statistically processing n pieces of browsing history information. good. For example, as shown in FIG. 13, when a plurality of routes of the barycentric coordinate positions of the distribution areas having the same start point are obtained as the route 1, the route 2, and the route 3, the browsing history information of each route May be used in correlation calculation as the final browsing history as the region information that appears most frequently. In the case of FIG. 13, since all the first browsing history information of each of the route 1 to route 3 is the block A02, the block A02 is adopted. Similarly, the second and third blocks A02 are used, but the fourth block A02, A05, and A03 appear one by one. In such a case, the region information that appears most frequently is regarded as indefinite and is deleted. Next, since the fifth and sixth blocks have two blocks A05 and one A03, the block A05 having a large number of appearances is adopted. Finally, the correlation calculating means 107 correlates the browsing history information as shown in FIG. Used for calculation. In this way, when there are multiple transitions of browsing locations starting from the same area, the correlation between the document structure blocks is calculated based on the transition of the browsing locations that occupy a larger number in the browsing history. The most general route is used for structuring, and there is an effect that it is possible to extract a more general document structure that is not easily affected by a particular individual's habit.

  In the description of the above embodiment, the correlation calculated by the correlation calculation unit 107 is added to the “continuation” score calculated by the structure recognition unit 102 to re-recognize the document structure, but is added to the “continuation” score. Instead of this, for example, it may be added to a score for which a value is set in common among highly correlated blocks. For example, when the correlation between the block 503 and the block 505 is calculated as +20 in the state where the score as shown in the upper part of FIG. 15 is calculated, a value is commonly set in the block 503 and the block 505. The “Continue with Paragraph” and “Continuation with 502” scores are incremented by +20, nothing is added to the “Continuation with 504” score, which is set only in the block 505, and updated to the score shown in the lower part of FIG. The document structure may be re-recognized.

  Next, a second embodiment of the present invention will be described in detail with reference to the drawings.

  Referring to FIG. 16, the document structuring apparatus 210 according to the second exemplary embodiment of the present invention is compared with the document structuring apparatus 110 according to the first exemplary embodiment shown in FIG. The difference is that a correlation calculation unit 212 is provided instead of the operation terminal 107, and an operation terminal 211, which is a document browsing terminal, is connected instead of the client terminal 105.

  The operation terminal 211 includes a mobile phone, a portable information terminal, a personal computer, and the like. The operation terminal 211 displays the document data distributed from the distribution unit 104 on the display device and presents it to the user, and responds to the user's operation. In addition to the function of requesting the distribution unit 104 to distribute the next document data, the correlation calculation unit 212 has a function of instructing the timing for calculating the correlation between the document structure blocks. The user of the operation terminal 211 browses the document data not for the purpose of browsing the document data but for the purpose of correcting the document structure information.

  The correlation calculation unit 212 is different from the correlation calculation unit 107 in FIG. 1 in that the correlation between the document structure blocks is calculated when receiving an instruction from the operation terminal 211, and the rest is the same as the correlation calculation unit 107. is there.

  Next, the overall operation of the present embodiment will be described by borrowing the flowchart of FIG.

  The document structuring apparatus 210 first reads out document data 101a obtained by digitizing a document page for recognizing the document structure from the document storage unit 101, and the structure recognizing unit 102 uses a known method in Patent Documents 1 to 3 and the like. The document data 101a is divided into several areas (document structure blocks), and the document structure (title, paragraph, etc.) of each area is recognized (step 201 in FIG. 2). The document structure information recognized by the structure recognition unit 102 is stored in the document structure storage unit 103.

  Next, the document structuring apparatus 210 refers to the recognized document structure information from the document structure storage unit 103 by the distribution unit 104 and distributes a part of the document data 101 a stored in the document storage unit 101 to the operation terminal 211. (Step 202). This data distribution is performed every time the user of the operation terminal 211 requests distribution of document data of an adjacent portion by scrolling the screen or the like. The user of the operation terminal 211 who browses the document data for the purpose of correcting the document structure information browses the document data so that the continuing document structure block is located at the approximate center of the display screen.

  Next, every time document data is distributed to the operation terminal 211 by the distribution unit 104, the document structuring apparatus 210 uses the browsing history acquisition unit 108 to record the history of browsing points by the user of the operation terminal 211 for the distributed document data. Is obtained by the method shown in FIG. 8, for example (step 203). The acquired browsing history is stored in the browsing history storage unit 106 as browsing history information 106a.

  Next, in the document structuring apparatus 210, at the timing when the start of correlation calculation is instructed from the operation terminal 211, the structure recognizing unit 102 uses the correlation calculation unit 212 based on the browsing history accumulated in the browsing history storage unit 106. The correlation between the recognized document structure blocks is calculated, and the calculation result is notified to the structure recognition means 102 (step 204).

  Next, the document structuring apparatus 210 re-recognizes the document structure of the document data 101 a by using the correlation calculated by the correlation calculation unit 212 by the structure recognition unit 102 and updates the document structure information 103 a in the document structure storage unit 103. (Step 205).

  Similarly, steps 202 to 205 may be repeated any number of times using the re-recognized document structure information.

  Next, the effect of this embodiment will be described.

  In this embodiment, since the operation terminal 211 can instruct the timing for calculating the correlation between the document structure blocks, the user of the operation terminal 211 can view the document data for the purpose of correcting the document structure information 103. When performing, correlation calculation can be performed at an appropriate timing.

  Further, in the present embodiment, even a person who does not have special knowledge or technology for creating a structure extraction rule used by the structure recognition unit 102 can simply browse a document through the operation terminal 211. There is an effect that the document structure information can be corrected.

  The browsing history acquisition unit 108 acquires the history of browsing locations by the user of the operation terminal 211 from the data distributed to the operation terminal 211 by the distribution unit 104 as in the first embodiment. The user directly designates a correlated document structure block in the browsing history acquisition unit 108, and the browsing history acquisition unit 108 stores the specified document structure block information in the browsing history storage unit 106 as browsing history information 106a. You may do it. In this case, the document structure can be re-recognized only by specifying the document structure block.

  Although the embodiments and examples of the present invention have been described above, the present invention is not limited to the above examples, and various other additions and modifications can be made. In addition, the document structuring apparatus of the present invention can be realized by a computer and a program as well as by realizing the functions of the document structuring apparatus as hardware. The program is provided by being recorded on a computer-readable recording medium such as a magnetic disk or a semiconductor memory, and is read by the computer at the time of starting up the computer, etc. And function as the structure recognition means 102, the distribution means 104, the correlation calculation means 107 or 212, and the browsing history acquisition means 108.

  The present invention can be applied to uses such as a document structure recognition device that recognizes a document structure from document data, a document distribution device that uses the recognized document structure, and a program for realizing these devices on a computer.

1 is a block diagram of a document structuring apparatus according to a first embodiment of the present invention. It is a flowchart which shows operation | movement of the document structuring apparatus concerning the 1st Embodiment of this invention. It is a flowchart which shows operation | movement of the structure recognition means in the document structuring apparatus concerning the 1st Embodiment of this invention. It is a figure which shows an example of the rule utilized in order for the structure recognition means in the document structuring apparatus concerning the 1st Embodiment of this invention to determine a score. It is a figure which shows the process of the structure recognition means in the document structuring apparatus concerning the 1st Embodiment of this invention. It is a figure which shows an example of the data format of the document structure information in the document structuring apparatus concerning the 1st Embodiment of this invention. It is explanatory drawing of the reference | standard for determining the image data which the delivery means in the document structuring apparatus concerning the 1st Embodiment of this invention delivers. It is a figure which shows an example of the browsing history information which the browsing history acquisition means in the document structuring apparatus concerning the 1st Embodiment of this invention acquires and accumulate | stores. It is a figure which shows the process in which the score which the structure recognition means computed by the correlation calculation means in the document structuring apparatus concerning the 1st Embodiment of this invention is updated, and the document structure is re-recognized. It is a figure which shows an example of the data format of the updated document structure information in the document structuring apparatus concerning the 1st Embodiment of this invention. It is a figure which shows another example of the browsing history information which the browsing history acquisition means in the document structuring apparatus concerning the 1st Embodiment of this invention acquires and accumulate | stores. It is a figure which shows another rule in which the correlation calculation means in the document structuring apparatus concerning the 1st Embodiment of this invention calculates a correlation. It is explanatory drawing of the statistical process of the browsing history by the correlation calculation means in the document structuring apparatus concerning the 1st Embodiment of this invention. It is a figure which shows an example of the browsing history statistically processed by the correlation calculation means in the document structuring apparatus concerning the 1st Embodiment of this invention. It is a figure which shows another procedure in which a structure recognition means updates a score using the correlation which the correlation calculation means calculated in the document structuring apparatus concerning the 1st Embodiment of this invention. It is a block diagram of the document structuring apparatus concerning the 2nd Embodiment of this invention.

Explanation of symbols

101 ... document storage unit 101a ... document data 102 ... structure recognition unit 103 ... document structure storage unit 103a ... document structure information 104 ... distribution unit 105 ... client terminal 106 ... browsing history storage unit 106a ... browsing history information 107 ... correlation calculation unit 108 ... browsing history acquisition means 109 ... processing unit 110 ... document structuring apparatus 201 ... document structure recognition step 202 ... document data distribution step 203 ... browsing history acquisition step 204 ... correlation calculation step 205 ... document structure re-recognition step 210 ... document structuring Device 211 ... Operation terminal 212 ... Correlation calculation means 301 ... Area division and character recognition step 302 ... Score calculation step 303 ... Document structure selection steps 501 to 505 ... Recognized blocks

Claims (19)

  Document storage means for storing document data, document structure storage means for storing document structure information of the document data, browsing history storage means for storing browsing history information, and document data stored in the document storage means as a document A history of the browsing position of the document data when distributed to the browsing terminal is acquired and stored as browsing history information in the browsing history storage means, and the document structure information of the document data is stored using the stored browsing history information. And a processing unit for re-recognizing the document structure of the document data using the calculated correlation result between the document structure blocks. Document storage means for storing document data;
Document structure storage means for storing document structure information;
Structure recognition means for recognizing a document structure from document data stored in the document storage means and storing it in the document structure storage means as document structure information;
Browsing history storage means for storing browsing history information;
Distribution means for distributing a part or all of the document data to a document browsing terminal with reference to the stored document structure information;
A browsing history acquisition unit that acquires a history of a browsing portion of a document distributed to the document browsing terminal and accumulates in the browsing history storage unit as browsing history information;
Correlation calculation means for calculating correlation between document structure blocks in the stored document structure information using the stored browsing history information;
The structure recognizing device, wherein the structure recognizing unit receives the correlation between the document structure blocks calculated by the correlation calculating unit and re-recognizes the document structure.   3. The document structuring apparatus according to claim 2, wherein the structure recognizing unit updates the accumulated document structure information based on a result of re-recognition.   The distribution unit distributes document data having a size corresponding to a display area of the document browsing terminal in one distribution, and the browsing history acquisition unit includes a document structure block included in the distributed document data. 4. The document structuring apparatus according to claim 2, wherein the information is acquired as a history of browsing locations of the document.   5. The document structuring apparatus according to claim 4, wherein the browsing history acquisition unit includes time information at which document data is distributed in browsing history information.   The correlation calculation means is configured to generate a document structure every time a certain number of the browsing history information is obtained, every time a certain period of time elapses, every instruction from the document browsing terminal, or any combination thereof. 6. The document structuring apparatus according to claim 2, wherein the document structuring apparatus calculates a correlation between blocks.   The correlation calculation means, when there are a plurality of browsing location transitions starting from the same document structure block, in the plurality of browsing history information related to the same document stored in the browsing history storage means, 4. The document structuring apparatus according to claim 2, wherein the correlation between the document structure blocks is calculated based on the transition of the browsing location occupied. a) a step of recognizing the document structure from the document data stored in the document storage means and storing the structure in the document structure storage means as document structure information;
b) a distribution means for distributing a part or all of the document data to a document browsing terminal with reference to the stored document structure information;
c) a browsing history acquisition unit that acquires a browsing history of a document distributed to the document browsing terminal, and stores the browsing history information as browsing history information in a browsing history storage unit;
d) a step of calculating a correlation between the document structure blocks in the stored document structure information using the stored browsing history information;
e) the structure recognizing means receiving the correlation between the document structure blocks calculated by the correlation calculating means and re-recognizing the document structure;
A document structuring method comprising: f) the structure recognizing means updating the stored document structure information based on the result of re-recognition;
The document structuring method according to claim 8, further comprising:   The distribution unit distributes document data having a size corresponding to the display area of the document browsing terminal in one distribution, and the browsing history acquisition unit stores document structure block information included in the distributed document data as a document. The document structuring method according to claim 8, wherein the document structuring method is acquired as a history of browsing locations.   11. The document structuring method according to claim 10, wherein the browsing history acquisition unit includes time information at which the document data is distributed in the browsing history information.   The correlation calculation means is configured to generate a document structure every time a certain number of the browsing history information is obtained, every time a certain period of time elapses, every instruction from the document browsing terminal, or any combination thereof. 12. The document structuring method according to claim 8, wherein the correlation between blocks is calculated.   The correlation calculation means, when there are a plurality of browsing location transitions starting from the same document structure block, in the plurality of browsing history information related to the same document stored in the browsing history storage means, 10. The document structuring method according to claim 8, wherein the correlation between the document structure blocks is calculated based on the transition of the browsed portion occupied. A computer having document storage means for storing document data, document structure storage means for storing document structure information, and browsing history storage means for storing browsing history information;
Structure recognition means for recognizing a document structure from document data stored in the document storage means and storing the document structure information in the document structure storage means;
Distribution means for distributing a part or all of the document data to a document browsing terminal with reference to the stored document structure information;
A browsing history acquisition unit that acquires a history of a browsing portion of a document distributed to the document browsing terminal and accumulates in the browsing history storage unit as browsing history information;
Correlation calculation means for calculating correlation between document structure blocks in the stored document structure information using the stored browsing history information;
The structure recognition means receives the correlation between the document structure blocks calculated by the correlation calculation means and re-recognizes the document structure.   15. The program according to claim 14, wherein the structure recognition unit updates the stored document structure information based on a result of re-recognition.   The distribution unit distributes document data having a size corresponding to a display area of the document browsing terminal in one distribution, and the browsing history acquisition unit includes a document structure block included in the distributed document data. 16. The program according to claim 14, wherein the information is acquired as a history of a browsing position of the document.   The program according to claim 16, wherein the browsing history acquisition unit includes time information at which document data is distributed in browsing history information.   The correlation calculation means is configured to generate a document structure every time a certain number of the browsing history information is obtained, every time a certain period of time elapses, every instruction from the document browsing terminal, or any combination thereof. 18. The program according to claim 14, 15, 16 or 17, which calculates a correlation between blocks.   The correlation calculation means, when there are a plurality of browsing location transitions starting from the same document structure block, in the plurality of browsing history information related to the same document stored in the browsing history storage means, 16. The program according to claim 14, wherein the correlation between the document structure blocks is calculated based on the transition of the browsing location occupied.

Images