JP5185089B2 - Content position estimation device - Google Patents

Description

  The present invention relates to a content position estimation apparatus that maps information such as content on a map.

The POI position on the map such as Web contents (news, blog, etc.) can be specified from the added GPS information, address, known (mapped) POI name, and the like. The identification of the position from the address is called geocoding and is performed based on the address database. Even in the absence of such information, attempts have been made to specify the position using various information. The technique disclosed in Non-Patent Document 1 below is a technique for specifying the position of a document such as a blog. In this technique, a character string of an address or a spot name included in a document is extracted from a character string that is strongly related to the subject matter in the document, and the most relevant one is estimated as a position on the map of the document. Patent Document 1 below discloses a technique for specifying a position on a map using address information and fragment of address information in a document. In the technique disclosed in Patent Document 1, in addition to address information, a telephone number, a landmark point name, an airport code, and the like are used.
Special Table 2007-524113 (summary) “Loco Sticker”, http://okilab.jp/project/location/

  However, when collecting content that does not know the geographical location or is described in an ambiguous expression and mapping it on the map, in the conventional mapping based only on the address or known spot name, the collected content is mapped to the map. There is a problem that it cannot be mapped on.

  In view of the above problems, an object of the present invention is to provide a content position estimation device capable of mapping content described in an ambiguous expression without address information or spot names on a map.

  In order to achieve the above object, according to the present invention, there is provided a content position estimation device for estimating a position to be originally mapped when mapping content including text on a map displayed on a display means. For each word of the text of one or more known contents whose position to be mapped is known, a first acquisition unit that acquires the content that is the target of power position estimation from the outside of the content position estimation device, A storage unit that stores map information including position information of an area specified as the position to be mapped of the known content of the included text; a dividing unit that divides the acquired text of the content into words; For each word, the map information including the position information of the area corresponding to the word is obtained from the storage means. Second acquisition means to be acquired next, generation means for combining the plurality of acquired map information to generate composite map information, and the position of the map to be mapped in the generated composite map information There is provided a content position estimation apparatus comprising: a determination unit that determines a position where the content on the map has the largest overlap is a position where the acquired content is to be mapped. With this configuration, content described in an ambiguous expression without address information or spot names can be mapped on a map. In addition, the display means here says the display etc. which a map built-in terminal etc. which are mentioned later have. The area position information described above is information such as coordinate data, for example.

  In the content position estimation apparatus of the present invention, the first acquisition unit acquires the known content, the division unit divides the acquired text of the known content for each word, and the generation unit is divided. It is a preferred aspect of the present invention to generate the map information for each word and store it in the storage means. With this configuration, map information for each word can be generated in advance.

  Further, in the content position estimation apparatus of the present invention, it is determined whether or not the acquired content and the known content are the same, and whether or not the position where the acquired content should be mapped can be specified. And when the acquired content and the known content are the same, the determining unit determines the position where the known content is to be mapped as the position where the acquired content is to be mapped. In addition, it is a preferable aspect of the present invention to determine the position to be mapped only when it is determined that the position to which the acquired content is to be mapped can be specified. With this configuration, the accuracy and efficiency of position estimation can be improved.

  In the content position estimation apparatus of the present invention, when the generation unit generates the composite map information, the acquired content acquisition site information, the content creator information, and the content viewer It is a preferable aspect of the present invention that the composite map information is generated in consideration of external information including at least one of the above information and information on the upper content of the category to which the content belongs. With this configuration, the accuracy and efficiency of position estimation can be improved.

  Further, in the content position estimation apparatus of the present invention, the generating means narrows down the mapping position area on the map using words and collocations indicating the positional relationship included in the acquired text of the content, and the composite map Generating information is a preferred aspect of the present invention. With this configuration, the accuracy and efficiency of position estimation can be improved.

  Further, in the content position estimation apparatus of the present invention, the generating means narrows down the area of the position to be mapped on the map based on the measured highly accurate position information, and generates the composite map information, This is a preferred embodiment of the present invention. With this configuration, the accuracy and efficiency of position estimation can be improved. Here, the position information corresponds to digital position information described later.

  Further, in the content position estimation apparatus of the present invention, when the acquired content is the posted blog information, the determining means sequentially determines a position to be mapped for each acquired blog information. This is a preferred embodiment of the present invention. With this configuration, the blog posting system can also cope.

  The content position estimation apparatus of the present invention has the above configuration, and can map content described in an ambiguous expression without address information or spot names on a map. In addition, by specifying the position of the content for which it is difficult to specify the position of the content, it is possible to increase the amount of content that can be handled by a map information site or the like. In addition, it is not necessary to specify address information, latitude / longitude information, etc. in order to specify the position of the content, and writing of a blog or the like can be made convenient.

  Embodiments of the present invention will be described. First, the configuration of the content position estimation apparatus according to the embodiment of the present invention will be described with reference to FIG. As shown in FIG. 1, the crawler 100 collects content through the Internet 108 and stores the collected content in the content DB 101. The position information analysis unit 102 acquires content stored in the content DB 101, and performs processing of a word distribution map generation phase and a content position estimation phase described later. The position information analysis unit 102 further includes a division unit 110, an acquisition unit 111, a generation unit 112, a determination unit 113, and a determination unit 114. The position information-added content DB 103 stores contents analyzed by the position information analysis unit 102 and provided with position information.

  The voice recognition system 104 recognizes a voice instruction of a user such as a car navigation (car navigation) system. The IME / input system 105 receives an instruction (command or the like) from a user or the like. The map server 106 stores map information. The built-in map terminal 107 has display means such as map information and a display, and displays on the display means such as a display map information in which the content whose content position has been estimated by the position information analysis unit 102 is mapped. . The map server 106 and the map built-in terminal 107 acquire the content whose position information is estimated and stored in the position information-added content DB 103 and map the content on the map. The high-accuracy position measuring device 109 measures high-accuracy position information such as GPS information.

  Here, an example of the word distribution map generation phase by the position information analysis unit 102 will be described with reference to FIGS. In this word distribution map generation phase, the geographical distribution of each word is obtained from content text whose position on the map is known in advance. The obtained geographical distribution for each word is used when the position of the content is estimated.

  As shown in FIG. 2A, content text whose position on the map is known is prepared in advance (step S201). Specifically, content texts 201 to 204 whose positions on the map (for example, Kamakura, Enoshima, Asakusa, etc.) as shown in FIG. The dividing unit 110 divides the prepared content texts 201 to 204 for each word (step S202). For this division, for example, a technique called morphological analysis (see http://chasen-legacy.sourceforge.jp/) is used. This method cuts out the morpheme, which is the smallest unit of Japanese meaningful word strings.

  After each content text is divided into words, the generation unit 111 obtains a geographical distribution 205 for each divided word (step S203). Here, an example of a method for obtaining the geographical distribution for each word will be described with reference to FIG. For example, when obtaining the geographical distribution in the word “sea”, it is confirmed which content text having the word “sea” corresponds to. In the case of the content text shown in FIG. 2B, the content texts 201 and 203 correspond to the content text having the word “sea”.

  The position indicated by each content text is “Kamakura” in the content text 201 and “Enoshima” in the content text 203. As a result, the grid corresponding to “Kamakura” and “Enoshima” is selected from the map 300 divided for each grid, and “1” is added to the selected grid. In the case of FIG. 3, numerical values are described in three grids, which means that a portion corresponding to the grid 301 has been selected twice. Thus, the geographical distribution (also referred to as a density map) for each word is obtained by expressing the density of each grid by a numerical value.

  Also, the point distribution can be very sparse, and there are aspects that are not sufficient for estimation. In that case, it is preferable to fill in a value (count) between points (or grids). Complementation may be performed by generating supplementary data at a known point or around the grid. As a method, a method of adopting an average of neighboring points or grids, a method of generating TIN (see Triangulated Irregular Network: http://en.wikipedia.org/wiki/TIN), a Kriging method (http: // www) .stat.math.keio.ac.jp / sympo / takeda.ppt.files / frame.htm). In this way, the geographical distribution 205 for each word shown in FIG.

  Next, an example of a content position estimation phase performed by the position information analysis unit 102 will be described with reference to FIGS. In this content position estimation phase, the location on the map of content whose address information is unknown is estimated using the geographical distribution generated in the word distribution map generation phase.

  First, a content text having no address information or the like (geographical position is unknown) is prepared (step S401). An example of the content text (target content text) prepared here is shown in FIG. As shown in FIG. 4B, the position of the content indicated by the target content text on the map is “unknown”, but the content content “A temple near the sea ...” is described.

  The dividing unit 110 divides such target content text into words (step S402). The acquisition unit 112 acquires the geographic distribution corresponding to the divided words from the geographic distribution generated in the word distribution map generation phase (the geographic distribution shown in FIG. 4C) (step S403). The generation unit 111 determines the combined distribution (the combined distribution shown in FIG. 4D) by combining the acquired geographic distributions (step S404). Here, the composite distribution is a combination of geographic distributions. In the case of the grid-like geographical distribution described above, the values for each grid may be simply added, but it is preferable to average (normalize) the number of words for each word.

  The determination unit 113 estimates the peak portion (high density portion) of the obtained composite distribution as the position of the content on the map (step S405). In the case of the composite distribution shown in FIG. 4D, the overlapping portion 400 is estimated as the position of the target content on the map.

  Here, in order to improve the accuracy and efficiency of estimating the position of the content on the map, after step S403 described above, it is determined whether the target content text is related to a known POI, and the target content text is It is also possible to determine whether or not it is related to POI. This determination is performed by the determination unit 114, for example. If the target content text is related to a known POI, it is not necessary to estimate the position of the content on the map, and the position of the known POI may be set as the position of the target content on the map. If the target content text is not related to the POI, it is not related to the POI in the first place, so there is no need to estimate the position on the map.

Whether the target content text is related to a known POI or not is determined by document unit proximity to the known POI content text (cosine distance in VSM (Vector Space Model) space using word structure) Or the like), and the content is determined to be the same content when it is equal to or greater than a predetermined threshold. The threshold is empirically determined from the correct data set using a learning method. VSM is used in determining the proximity between documents, vector V by different number of words w of each document is first _{(V = (w 1, w} 2, ···, w i, ···, w n)) Define In this case, the proximity is represented by an inner product of vectors (sim (D _i , D _j ) = V _i · V _j ). As a case where it is clear that the documents represent the same POI, it is determined whether or not the inner product is a predetermined value or more. The details of the VSM described above are described in the following document (URL). http://unicorn.ike.tottori-u.ac.jp/murakami/paper/INTERNATIONAL/PACLING_2001_9/murakami/

  Examples of a method for determining whether the target content text is related to POI include the following. This method is to determine that the document is not related to POI when the variance of the composition distribution by words included in the target content text exceeds a certain threshold value (when it is too scattered and cannot be specified).

  Furthermore, in order to improve the accuracy and efficiency of location estimation on the map of the content, not only the words of the content text, but also the site (there may be locality in each site), the writer (location is the writer's It is possible to estimate the location of the content on the map using external information such as the viewer, the title of the higher-level content (the location may be specified to some extent by category, etc.) is there. Furthermore, it is possible to superimpose the appearance frequency for each word and the estimated position obtained from the information such as the above-mentioned viewer to correct the information on the geographical distribution of the word. As a result, it is possible to suppress the influence of words or the like that appear in the content by chance although they are not related.

  Furthermore, in order to improve the accuracy and efficiency of estimating the position of the content on the map, the position of the target content on the map is determined from the word indicating the positional relationship with the known POI using words and collocations indicating the positional relationship. It is also possible to estimate. Transform positional relationships from geometric (right, next, beyond, back, foreground, etc.) and geographic (south, foot, coast, etc.) words into spatial operators that represent the actual relationships and express them with those operators Narrow down the range. A range based on a geometrical and spatial positional relationship for each operator is also represented by a spatial probability distribution.

  A specific range of focusing will be described. The document “The store is in the south of the temple” can be translated into an operator with predicate and argument. That is, it can be converted into a predicate logic or functional language (Lisp) style description such as (predicate, arg1, arg2,...) → (south, shop, temple). It is also possible to apply this to logical reasoning (inferring where the store is located). Based on this result, a geographical range corresponding to the operator is designated. Further, when there is information such as (next to the store, school), (alongside, store, river), it is possible to narrow down geographically as shown in FIG. The range may be narrowed down also using geographical shape information and attribute information indicating the shape of terrain, buildings, roads, water bodies and the like. The attribute information refers to information on the use of a building or land.

  Furthermore, in order to improve the accuracy and efficiency of position estimation on the map of content, high-precision digital position information (probe car, portable access log) such as GPS information measured using the high-precision position measurement device 109 Based on a set (distribution) of trajectory information based on images with GPS information, etc., and articles posted to GPS (not related to the target content), narrowing down to only POIs that are human action targets Is also possible. As a result, it is possible to eliminate locations where there is no POI such as empty mountains, outside roads, fields, etc., and to place more weight on places where people often visit, so make more accurate estimations. Can do.

  The above-described method can also be used in a system in which a user inputs text information and posts it (such as a blog posting system), and the position on the map is sequentially determined based on the currently input text. It is also possible to estimate and present the position. Based on the position information, for example, the map beside the text input screen can be scrolled or zoomed.

  In the above-described method, the position on the map can be estimated from the content of the text already input during the voice recognition in the system that can recognize the voice from the user such as a car navigation system, and the voice recognition based on the position information. It is also possible to perform tuning in real time. For example, when keywords such as “Sea” and “Surfing” appear, the estimated probability of occurrence utterances by topic is increased (“Southern”, etc.) and related place names such as “Shonan”, “Kujukurihama”, etc. Increase the estimated probability value. Furthermore, the estimated probability value of a word that is not directly related to the text but is related via a geographical position (such as “Enoden”) is also increased.

  Next, an example of the processing flow of the content position estimation apparatus according to the embodiment of the present invention will be described with reference to FIG. Note that the processing entity of each step is a content position estimation device. As shown in FIG. 6, first, content text (content text whose geographical position is unknown) of content whose address information is unknown and whose position on the map cannot be specified is acquired (step S601). The acquired content text is set as the target content text, and the target content text is divided into words (step S602). The geographical distribution (density map) for each divided word is acquired (step S603).

  Here, based on the acquired geographical distribution, it is determined whether or not the target content is the same as the known content (whether it matches the known POI: step S604). Is output (step S605). If they are not the same, the obtained geographic distribution is synthesized to obtain a synthesized distribution (step S606). Here, it is determined whether or not the distribution of the composite distribution exceeds a predetermined threshold (is not content related to POI: step S607), and if it exceeds, the process ends without outputting a position. If not exceeded, the estimated position obtained from the external information described above is superimposed on the composite distribution (step S608).

  Distributions are generated from words indicating positional relationships, and the generated distributions are further superimposed (step S609). Using the digital position information described above, the candidate locations for POI are narrowed down (step S610). Then, the final peak of the composite distribution is estimated as the position of the target content on the map (step S611). Steps S604, S605, S607, and S608 to S610 are options for improving the accuracy and efficiency of estimating the position of the content on the map, and are not necessarily required, but there are these steps. Is a preferred embodiment.

  Since the content position estimation apparatus according to the present invention can map content described in an ambiguous expression without address information or spot names on a map, the content position estimation apparatus maps information such as content on the map. It is useful for such as.

It is a figure including the block diagram which shows an example of a structure of the content position estimation apparatus which concerns on embodiment of this invention. It is a flowchart which shows an example of the step of the word distribution map production | generation phase in embodiment of this invention. It is a figure which shows an example of the text of the content (known content) by which the position on the map in embodiment of this invention is known. It is a figure which shows an example of the geographical distribution for every word in embodiment of this invention. It is a figure for demonstrating an example of the method of calculating | requiring the geographical distribution for every word in embodiment of this invention. It is a flowchart which shows an example of the step of the content position estimation phase in embodiment of this invention. It is a figure which shows an example of the object content text in embodiment of this invention. It is a figure which shows an example of the geographical distribution produced | generated in the word distribution map production | generation phase in embodiment of this invention. It is a figure which shows an example of the synthetic | combination distribution in embodiment of this invention. It is a figure for demonstrating the aspect in the case of narrowing down geographically in embodiment of this invention. It is a flowchart for demonstrating an example of the content position estimation apparatus processing flow which concerns on embodiment of this invention.

Explanation of symbols

100 crawler (first acquisition means)
101 Content DB
102 position information analysis unit 103 content DB with position information (storage means)
DESCRIPTION OF SYMBOLS 104 Speech recognition system 105 IME, input system 106 Map server 107 Built-in map terminal 108 Internet 109 High-precision position measuring device 110 Dividing unit (dividing means)
111 Generator (Generator)
112 Acquisition Unit (Second Acquisition Unit)
113 Determination unit (determination means)
114 Judgment Unit (Judgment Means)
201, 202, 203, 204 Content text 205 Geographic distribution for each word 300 Map 301 Grid 400 Overlapping part

Claims (7)

A content position estimation device that estimates a position to be originally mapped when mapping content including text on a map displayed on a display means,
First acquisition means for acquiring content whose geographical position to be estimated is unknown from outside the content position estimation device;
For each word of text of one or more known contents whose position to be mapped is known, geographical distribution information including position information of the area specified as the position to be mapped of the known content of the text including the word is stored Storage means to
Dividing means for dividing the acquired text of the content into words;
Second acquisition means for sequentially acquiring the geographical distribution information including position information of the area corresponding to the divided words from the storage means for each word;
Generating means for combining the plurality of acquired geographical distribution information to generate combined distribution information;
In generated the synthesized distribution information, and determination means for overlapping on the map location to be pre-Symbol mapping to determine the most frequently positions to be mapped in the acquired content location,
A content position estimation apparatus provided. The first acquisition means acquires the known content,
The dividing unit divides the acquired text of the known content into words,
The content position estimation apparatus according to claim 1, wherein the generation unit generates the geographical distribution information for each of the divided words and stores the generated information in the storage unit. A judgment means for judging whether or not the acquired content and the known content are the same, and determining whether or not a position to be mapped of the acquired content can be specified;
When the acquired content and the known content are the same, the determination unit determines a position where the known content is to be mapped as a position where the acquired content is to be mapped, and the acquired content is The content position estimation apparatus according to claim 1 or 2, wherein the position to be mapped is determined only when it is determined that the position to which the content should be mapped can be specified. When generating the composite distribution information, the generating means acquires the content acquisition site information, the content creator information, the content viewer information, and the upper content of the category to which the content belongs. The content position estimation apparatus according to any one of claims 1 to 3, wherein the composite distribution information is generated in consideration of external information including at least one of the information. The said generation | production means narrows down the area of the mapping position on a map using the word which shows the positional relationship contained in the acquired text of the said content, and a collocation, and produces | generates the said synthetic | combination distribution information. The content position estimation apparatus according to claim 1. The content according to any one of claims 1 to 5, wherein the generation unit narrows down an area of a position to be mapped on a map based on the measured highly accurate position information, and generates the composite distribution information. Position estimation device. When the acquired content is blog information posted,
The content position estimation apparatus according to any one of claims 1 to 6, wherein the determination unit sequentially determines a position to be mapped for each acquired blog information.

Images