JP5404134B2 - Label placement device - Google Patents

Description

  The present invention relates to a label placement device that places a label indicating a point attribute on a map.

  In general map services, it is basic to present a POI (Point of Interest) with a label such as a point + place name. The contents beyond that are individually presented, for example, when the user clicks and presents them in a pop-up. When a large amount of POI is a target, it is very complicated. In principle, text information and the like are not included when the map is first viewed, and therefore, searching and browsing operations in several steps are necessary to grasp the contents. On the other hand, in GIS (Geographic Information System), etc., quantified attributes are presented as point distributions, symbols are displayed at the aggregated representative points, and contours are drawn on the surface that complements them. There are various ways.

  Although the contents to be displayed are various, GIS is a method for presenting a quantified value assigned to a feature or an area. Although it is possible to present a distribution of multiple attributes at once, there is a tendency that even if only a few distributions are displayed, it becomes a complicated figure that is difficult to decipher. On the other hand, if each label is arranged only in a prominent place for each attribute, this problem can be dealt with. However, a simple method may be used when the same label appears only at one point, such as a place name, but there are many such attributes as described above, and they cannot always be displayed in one place. In such a case (when the distribution is scattered to some extent), the number of labels tends to be too large and needs to be dealt with. In other words, since there are many attributes to be displayed, it is desired to present only representative points. However, when one representative point cannot represent the distribution of the attributes, special considerations that are not conventionally required are necessary.

  For example, attributes (labels) such as “Tokyo” and “Kanagawa” may be expressed by a single representative point, but “Rias Coast”, “Sunrise Spot”, “Scenic Spot”, “Tourist Boat”, “ Labels such as “pine group fabric” do not appear in one place because the distribution is scattered, and there are innumerable attributes, so if you try to represent them simultaneously, it becomes a very complicated and difficult to decipher map. Further, since the distributions overlap, there is a high possibility that one label will compete with the other label. There is a need for an adjusting means that allows these to be easily read by the user and that does not impair the actual distribution information. In addition, when a single attribute appears only in one place like a place name, the method disclosed in Patent Document 1 below is used as an example.

  In order to solve the above-described problems, a method is conceivable in which a rough label for each area can be grasped in advance on a map by displaying a label corresponding to the area in some manner. There are the following methods for this. The first method is a method of presenting a map for each predetermined attribute (label). The second method is a method for presenting a representative label on a map. The third method is a method in which areas are first explicitly divided on a map, and a label is presented to each area.

  The first method only presents the distribution for each attribute, but when the contents are text, it is impossible to display all the attributes and grasp the contents. The distribution can also be displayed on the map by limiting the label, but this can be done by using existing contour drawing or the like. The operation is complicated because a plurality of map displays are switched. The second method can be realized relatively easily by arranging a label at the highest point of the contour line of each attribute (text) obtained by the first method. Compared to the first method, it is superior in that it can be grasped on one screen. However, the visibility may deteriorate if the labels overlap.

  The third method is to divide areas in advance on a map and arrange labels. Thereby, the visibility of the label is improved. In the third method, when dividing an area based on the distribution of scattered points (including representative points of surfaces and areas) data (for example, data of features, observation points, etc.) on the map of point data Clustering (grouping) based on the proximity (distance) of the data and the proximity of the attribute of the point data (for example, similar types of stores, similar vegetation, similar generation distribution, similar word of mouth, guide text, etc.) Reference 2).

  According to the technique disclosed in Patent Document 2, clustering based on both (proximity on the map and proximity of the attribute) is a uniform content based on the content, and is gathered to some extent on the map An area (which is easy to see visually) is obtained. This technology can be applied to applications that create simple summary maps on maps.

Japanese Patent Laid-Open No. 9-237034 (summary) JP 2008-224993 A (summary)

  However, both the second and third methods have problems. That is, depending on the data to be handled, it is not always possible to avoid overlapping labels when label placement is performed, and optimal label placement may not be achieved.

  An object of this invention is to provide the label arrangement | positioning apparatus which can perform the label arrangement | positioning in consideration of the appearance from a user, easiness of recognition, and usability in view of said problem.

  In order to achieve the above object, according to the present invention, storage means for storing position information on a map of each of a plurality of points and label data indicating the attribute of the point for each point, and storing in the storage unit A plurality of labels for each of the labels arranged by the arrangement means, and arrangement means for arranging the labels at the positions of the points on the map on the basis of the position information and the labels. Judging means for judging whether or not a predetermined condition for placement is satisfied, and an adjusting means for adjusting the placement of the placed label when the judging means judges that the predetermined condition is not satisfied. A label placement device is provided. With this configuration, it is possible to arrange labels in consideration of appearance from the user, ease of recognition, and ease of use. Here, the adjusting means may adjust the label arrangement until all labels satisfy a predetermined condition, or may stop the label arrangement adjustment to some extent. The label data here may be text data or data indicating a symbol or the like.

  Moreover, in the label arrangement apparatus of the present invention, it is a preferable aspect of the present invention that the adjustment unit adjusts the label arrangement by performing any one of label deletion, integration, and movement. . With this configuration, visibility on a map can be improved.

  Further, in the label placement device of the present invention, the adjustment means is within a range where the predetermined visibility of the label on the map can be ensured, and each label placed by the placement means on the map. It is a preferable aspect of the present invention to adjust the arrangement of the label so as to optimize the accuracy of the arrangement position. With this configuration, both visibility and fidelity can be reflected.

  Further, in the label placement device of the present invention, when the placement unit does not place the label for each point, a predetermined clustering process is performed based on the position information and the label stored in the storage unit, The label is arranged for each cluster, and the adjusting means is within a range in which a predetermined goodness on the map of each cluster arranged by the arranging means can be secured. It is a preferable aspect of the present invention to adjust the arrangement of the clusters so that the accuracy of the contents is optimized. With this configuration, both visibility and fidelity can be reflected.

  In the label placement apparatus of the present invention, when the adjustment unit calculates the optimization parameter, the adjustment unit estimates the parameter with an initial value of predetermined information indicating the complexity of the label placement or the appearance of the cluster. This is a preferred embodiment of the present invention. With this configuration, a plurality of parameter calculation processes can be reduced.

  Further, in the label placement device of the present invention, the adjustment means is on the map within a range in which a predetermined accuracy of the placement position of each label placed by the placement means on the map can be secured. It is a preferable aspect of the present invention to adjust the arrangement of the label so as to optimize the visibility of the label. With this configuration, it is possible to extract an attribute group that is geographically related while suppressing a deviation in accuracy to some extent.

  Further, in the label placement device of the present invention, when the placement unit does not place the label for each point, a predetermined clustering process is performed based on the position information and the label stored in the storage unit, The label is arranged for each cluster, and the adjustment means is on the map of each cluster arranged by the arrangement means within a range in which a predetermined accuracy of the contents of the cluster on the map can be ensured. It is a preferable aspect of the present invention to adjust the arrangement of the clusters so as to optimize the unity. With this configuration, it is possible to extract an attribute group that is geographically related while suppressing a deviation in accuracy to some extent.

  The label arrangement apparatus of the present invention has the above-described configuration, and can perform label arrangement in consideration of appearance from the user, ease of recognition, and ease of use.

It is a block diagram which shows an example of a structure of the label arrangement | positioning apparatus which concerns on embodiment of this invention. It is a figure for demonstrating grouping the some attribute in embodiment of this invention into several attributes. It is a figure for demonstrating the method of calculating | requiring a label position separately from the separate attribute distribution in embodiment of this invention. It is a figure for demonstrating the place comprised using spatial complementation about each attribute in embodiment of this invention. It is a figure for demonstrating the determination method of the label position through the area in embodiment of this invention. It is a figure for demonstrating the method to obtain a label from the attribute of a member after performing area division in embodiment of this invention. It is a figure which shows the mode of the geographical clustering in embodiment of this invention. It is a figure which shows the mode of the semantic clustering in embodiment of this invention. It is a figure which shows the mode of the geographical meaning clustering in embodiment of this invention. It is a figure which shows the relationship of the nearest label between different types of labels in label A and B in embodiment of this invention. It is a figure which shows the relationship of the nearest label between the labels of the same kind in label A and B in embodiment of this invention. It is a figure for demonstrating an example of the rearrangement of the label in embodiment of this invention. It is another figure for demonstrating an example of the rearrangement of the label in embodiment of this invention. It is a figure which shows the Delaunay diagram for demonstrating the parameter optimization method in embodiment of this invention. It is a figure which shows the concept of the trade-off between the goodness of a cluster and a fidelity in embodiment of this invention. It is a figure for demonstrating the example of parameter optimization in embodiment of this invention. It is a flowchart which shows an example of the processing flow of parameter optimization in embodiment of this invention.

  An example of the configuration of the label placement apparatus according to the embodiment of the present invention will be described with reference to FIG. The label placement apparatus includes a map drawing unit 101, a map drawing engine 102, a map DB (Data Base) 103, a label placement unit 104, a label extraction unit 105, a POI DB 106, a clustering function unit 107, and an area extraction unit 108. . Note that the clustering function unit 107 and the area extraction unit 108 are necessary components when performing clustering processing to be described later. The map drawing unit 101 displays map information retrieved from the map DB 103 by the map drawing engine 102 on a display unit (not shown) (not shown). Further, the map drawing unit 101 displays labels rearranged by a label arranging unit 104 described later on the map. The map drawing engine 102 searches for and acquires desired map information from the map DB 103. The map DB 103 stores map information.

  The POI DB 106 is a database that stores position information and attribute information (attribute information) related to the POI for each POI. Here, the attribute information is not limited to one for one POI, and there may be a plurality of attribute information. The label extraction unit 105 acquires position information and attribute information related to the POI from the POI DB 106, and arranges POI attributes (labels) at corresponding positions on the map. The POI attribute group arranged here is set as a labeling candidate. In the case of a keyword attribute, the keyword can be used as a label as it is. For other attributes, numerical values are used, and the numerical values are mapped according to the size and color of symbols on the display.

  Further, the label extraction unit 105 acquires position information and attribute information related to the POI from the POI DB 106, and does not place the POI attributes at the corresponding positions on the map, as shown in FIG. (A1 to An) may be arranged in a number of attributes (B1 to Bm). In this case, a new (reduced number) attribute set is defined using principal component analysis, clustering, or the like. The new attribute value must reflect the value of the original attribute.

  For example, in the case of a keyword vector space, the process of collecting frequently co-occurring keywords based on the number of keywords existing for each POI (document) is used by using Latent Semantic Indexing (SVD (Singular Value Decomposition)) The keyword vector dimension reduction method). The degenerated keyword vector is a new attribute and is a linear combination of the original keyword attributes. Therefore, the original keyword with high weight can be adopted as the representative label.

  The label placement unit 104 performs final label placement (rearrangement or placement adjustment) by performing predetermined processing on the labeling candidates. Note that the label placement unit 104 may perform the label placement process and the rearrangement process only by the label extraction unit 105 extracting the label. Further, when the labels are rearranged, the labels may be rearranged after determining whether or not the label arranging unit 104 satisfies a predetermined condition (such as a threshold value t described later). Other components may be determined.

  Here, a label arrangement method will be described. In order to arrange the labels on the map, the labels are arranged from the POI attribute data based on the geographical distribution. A single attribute (for example, a house, a factory, or the like) or a composite attribute that combines a plurality of attributes (such as an attribute that combines a plurality of keywords) may be used.

  A method for obtaining the label position separately from individual attribute distributions will be described. The label position can be obtained directly from the attribute and coordinate data. As a method, for example, a label is arranged at the maximum point of the distribution. Specifically, a label is arranged on the POI where the attribute has the maximum value (or maximum value). There may be multiple representative points. As another method, for example, a label is arranged at the center of the distribution. Specifically, it is arranged at the center of gravity of each POI. A weighted centroid based on the value of the POI attribute may be taken. In this method, the label position can be easily determined. This is shown in FIG.

  In the above case, when only a single POI has an attribute (for example, blue) of 0 or more, it becomes a summary label of a single POI. If the attribute is 0 or more for a plurality of attributes (for example, red and blue), the POI is counted in both red and blue distributions. Further, when the text is based on a vector space model (count vector for each word), the number of attributes is very large, such as tens of thousands, and it is necessary to present only the prominent ones among the attributes. In such a case, it is necessary to narrow down candidate attributes (words) in advance by a document summarization technique (for example, word importance calculation by TF / IDF). Further, based on this importance, it can be made a part of conditions for label deletion and integration at the time of label rearrangement described later.

  Next, a method for determining an attribute area and determining a label position will be described. The data from which the label is generated is a point distribution in which each point has a plurality of attributes. Assuming that each attribute has a scalar value, a field as shown in FIG. 4 can be configured using spatial interpolation for each attribute (here, red and blue attributes). As a complementing method, there are a method of obtaining a contour line by configuring a TIN (Triangulated Irregular Network), a method using linear and spline complementation, and the like. If the contour line is drawn with the Z value in the obtained field, an area as shown in FIG. 4 is obtained. However, the areas are not necessarily mutually exclusive as shown in FIG.

  For example, the label position can be defined by arranging a label near the local maximum point of this field. There are the following methods for determining the label position. Although it is almost the same as the case of obtaining from a point distribution, by defining an exclusive area, it is possible to exclude overlapping areas (mixed attributes) and place an area where only certain attributes stand out as a representative point There are merits such as being able to take.

  As the method, for example, a label is arranged at the maximum point of the distribution. Specifically, a point (point) larger than the surrounding is selected from the points. A threshold value is set, and a particularly high value is selected from the maximum points (see A1 in FIG. 5). As another method, for example, a label is arranged at the center of the distribution. Specifically, it is arranged at the center of gravity of a distribution polygon that is equal to or greater than a certain threshold (see A2 in FIG. 5). As another method, for example, a label is arranged at the center of the exclusive area. Specifically, a label is arranged at the center of an area where only the attribute value exists (see A3 in FIG. 5). As an application, it is possible to perform clustering on surface (polygon) data such as administrative divisions (reducing point clustering by using representative points).

  Next, a method of obtaining a label from member attributes after performing area division will be described. As a method for obtaining the above-described area, there is a method for obtaining a target label by summarization within a cluster set by performing geographic semantic clustering as in the technique disclosed in Patent Document 2. In the above method, the attribute (or composite attribute) is selected in advance. However, in the method disclosed in Patent Document 2, the area is first clustered in consideration of both the attribute (meaning) and the geography (latitude / longitude coordinates). Obtain a representative attribute (label) in a predetermined area. Eventually, the application shows the area and its label (see FIG. 6).

  Geographical semantic clustering is performed by adding a map space vector (latitude and longitude) to an attribute (eg, word space) vector. By integrating the word space vector and the map space vector, semantically close POIs are preferentially selected from the geographically close POIs and clustered. As a clustering method, a general clustering method may be used, such as an SOM, a k-means method, and an EM algorithm. The method of synthesizing a geographic vector and a semantic vector is a method of simply weighting the other vector, distance (similarity) (Euclidean distance, cosine similarity, etc.) There is a method in which space and semantic space are defined separately and the similarity is weighted. These weights are used as geographical semantic ratios.

  When clustering is performed using only geographical attributes (see FIG. 7A), clustering is simply performed based on geographical proximity, and homogeneity for each area is not guaranteed. On the other hand, when semantic clustering is performed (see FIG. 7B), it looks very difficult to understand (a lot of enclaves and the like), and the appearance is sacrificed. Therefore, by performing a compromise between the two and performing clustering that takes both into account, an area that is well-organized and is as homogeneous as possible can be obtained (see FIG. 7C). It is considered that the optimal geographical semantic clustering can be obtained by adjusting the geographical semantic ratio.

  Next, a label arrangement changing method will be described. As described above, the distribution is rarely concentrated in one place, and there are usually a plurality of extreme values. If the label display is simply performed by the method described above, the entity is displayed faithfully, but it tends to be complicated. Only the label positions obtained by the method described above are shown in FIG. 8A shows the relationship of the nearest neighbor label between different types of labels in labels A and B, and FIG. 8B shows the relationship of the nearest neighbor label between labels of the same kind in labels A and B. . In this case, there is a problem that labels are too close to each other to be discriminated and there are redundant labels. Here, not only the label obtained from the point attribute data but also the overlap with the label of the basic map serving as the ground can be considered in the same manner (the label set C may be introduced).

  As a method for solving the above-described problem (using the above-described label placement unit 104), there is the following. The first is a method of deleting one of the nearest points of different types of labels. Specifically, when the distance between the nearest labels A and B is equal to or less than the threshold value t, only the ones with large values of A and B (scalar values of predetermined attributes) are left (for example, FIG. Delete one of the circles in 8 (a)). The second method is to keep the nearest points of different types of labels away from each other. Specifically, when the distance between the nearest neighbors A and B is equal to or less than the threshold value t, the distance between A and B is set to be t (see FIG. 9A). The third is a method of merging with the nearest label. Specifically, if there is no other type of label between the nearest type of the same type of label, a label having a distance between the same type of labels having a threshold value t or less is merged, and a new label is added at the center point between both labels. Deploy. An appropriate threshold value may be set by combining indices described later.

  Here, there are some indicators of the importance of the labels for setting the threshold t when deleting and integrating the labels, and the contents for the indicators are as follows. For example, there is an inter-species distance (t1) as an index, and it is determined how far away from the different types of labels. The farther away the label, the harder it is to delete. In addition, there is an inter-kind distance (t2) as an index, and it is determined how far away from the same kind of label. The distant labels are less integrated. In addition, there is the same kind representative degree (t3) as an index, and it is determined by the label representing the label kind. The more representative the label type, the harder it is to delete. For example, if the number is 1 / label, the representativeness increases as the number of labels of the same type decreases.

  In addition, there is a label altitude (t4) as an index, and when the label position is determined by an extreme value-based method, it is determined by its Z coordinate (altitude). The higher the value, the harder it is to delete. In addition, there is a label influence range (t5) as an index, and when the label position is determined by an area-based method, it is determined by the size of the original area. The wider it is, the harder it is to delete. In addition, there is a label importance (t6) as an index, and when the importance is given to the label type itself, it is determined based on the importance. The higher the importance, the harder it is to delete. In addition, there is a label overlapping degree (t7) as an index, and it is determined based on a similarity with a label in the vicinity (such as within a radius t). The higher the similarity, the easier it is to delete.

Appropriate threshold values are set using values obtained by combining some or all of these indices by linear combination in consideration of polarity (whether or not easy to be deleted). In this case, the composite index is t ^opt .

  Next, a parameter optimization method will be described. The determination of the optimum threshold t is greatly influenced by the combination of attributes to be displayed, the display location, and the like, and thus cannot be uniquely determined. The complexity of the result display is used as an index for optimizing t according to the data to be displayed. The complexity of the result display and the fidelity of the content display are often a trade-off. That is, since t is increased and labels that are relatively far apart are merged and deleted, complexity can be suppressed, but fidelity is lost. Conversely, if t is reduced and only the labels that are in close proximity are merged and deleted, the complexity remains large but the fidelity is maintained. Also in this case, an optimum value is set by adjusting a parameter corresponding to the geographical semantic ratio in the area dividing method described above.

  Here, as a complexity index, a set of labels can be regarded as a cluster, and an index for good cluster (a CH index described later) can be used. In addition, when a network such as a network indicated by a solid line or a broken line in FIG. 9 is used, a neighbor graph (such as a Delaunay diagram (FIG. 10)) is created, and an index using the adjacent relationship of similar labels on the graph can also be used. For example, in FIG. 9B, links connecting B between labels A intersect each other, and by counting the number of such intersections, the distribution of label arrangement and the complexity of appearance can be defined.

  On the other hand, when an area is generated and displayed using the method of obtaining a label from the member attribute after performing the area division described above, it is necessary to optimize the area display. This method optimizes while changing the geographical semantic ratio. At this time, a threshold is set for the goodness of the cluster such as CH (Calinski Harabasz) index, spatial autocorrelation, etc. (there is little variation in the cluster and the distance from other clusters is far), and the fidelity (proximity by attribute) The degree of adoption). The conceptual diagram is shown in FIG.

Here, an example of parameter optimization will be described. When actual data is used in the above steps, the following CH ratio changes can be seen in the geographical semantic ratio (see FIG. 12). An easy-to-read area can be obtained by setting a threshold value based on the CH ratio. See below for CH index.
See http://www.msi.co.jp/vmstudio/materials/tech/cluster.html.

  Therefore, the CH index is an indicator that the variation between clusters is good and the proximity within the cluster is high, and is used as an indicator of good or bad clustering results. In addition, similar effects may be obtained with Hartigan Validation, Ratkowsky and Lance Validation, Dunn ’s Validity Index, and the like. When presenting a label directly without going through a cluster (area), the same type of label may be regarded as a cluster.

  If the optimum value obtained above has a strong correlation with the initial value of complexity (for example, the CH index when TH = 0 (no position relocation is performed)) on a given data set. The relationship between the optimum value and the initial value is obtained, and the optimum value can be uniquely estimated without performing a plurality of trials.

  Here, an example of the parameter optimization processing flow is shown in FIG. An optimization method for both label deletion and clustering parameters is shown. An initial value that maximizes complexity (a state in which no label is deleted in the case of label deletion: a threshold value 0, a ratio that considers only a semantic attribute in the case of clustering) is set as an initial value. Then, gradually change the parameters in a direction that reduces the complexity, and repeat until the complexity becomes acceptable. The time when the complexity becomes acceptable for the first time is the optimum value (optimum parameter).

  From the above, not only the geographical distribution but also the dimension to be visualized can be targeted. In the same way, a set of dimensions that should be visualized or divided into multidimensional data (in the above, geographical coordinates, images, coordinates of the text (2, 3 dimensions), time series observation data) This can be applied to the case of performing space-attribute clustering with a time (one-dimensional, etc.). For example, observed biological composition, clustering of ages based on historical events, and the case of clustering time series of human behavior.

  The attribute relating to the point (POI) can be applied not only to text but also to the following attribute values and combinations thereof. If the clustering can be performed appropriately, it may not be a scalar amount (genre etc.). For example, there are attributes of features (genre, population, sales, land prices, vegetation, mineral resource distribution, geographically related multimedia information such as voice, image, video, etc. and their characteristic quantities). Further, not only points but also lines (such as roads), planes (such as administrative divisions), and solids (such as buildings) can be defined by defining representative points. In addition, there are characters on screens, magazines, photos, blocks, windows, and time point attributes (time series measurement data, chronology).

  Since the label placement device according to the present invention can perform label placement in consideration of appearance from the user, ease of recognition, and ease of use, a label placement device that places a label indicating a point attribute on a map It is useful for such as.

101 Map Drawing Unit 102 Map Engine Unit 103 Map DB
104 Label placement unit (judgment means, adjustment means)
105 Label extraction unit (placement means)
106 POI DB (storage means)
107 Clustering function unit 108 Area extraction unit

Claims (7)

Storage means for storing, for each point, position data on a map of each of a plurality of points and label data indicating the attributes of the points
Based on the position information and the label stored in the storage means, an arrangement means for arranging the label for each point at the position of the point on the map;
For each of the label that is arranged by the arrangement means, a predetermined condition for multiple label placement, determine whether they meet the predetermined condition that the distance between said labels is greater than a predetermined threshold value A judgment means to
An adjusting means for adjusting the placement of the placed label when the judgment means judges that the predetermined condition is not satisfied,
A label placement device.   The label placement apparatus according to claim 1, wherein the adjustment unit performs placement adjustment of the label by performing any one of label deletion, integration, and movement.   The adjustment means optimizes the accuracy of the arrangement position of each label arranged by the arrangement means on the map within a range in which the predetermined visibility of the label on the map can be ensured. The label placement apparatus according to claim 1, wherein the placement of the label is adjusted. When the placement unit does not place the label for each point, the placement unit performs a predetermined clustering process based on the position information and the label stored in the storage unit, and places the label for each cluster.
The adjusting means optimizes the accuracy of the contents of the clusters on the map within a range in which the predetermined unity on the map of each cluster arranged by the arranging means can be secured. The label placement apparatus according to claim 1, wherein the placement of the clusters is adjusted.   The label arrangement according to claim 3 or 4, wherein the adjustment unit estimates the parameter with an initial value of predetermined information indicating a label arrangement or a cluster appearance complexity when calculating the optimization parameter. apparatus.   The adjustment means optimizes the visibility of the label on the map within a range in which a predetermined accuracy of the arrangement position on the map of each label arranged by the arrangement means can be secured. The label placement apparatus according to claim 1, wherein the placement of the label is adjusted. When the placement unit does not place the label for each point, the placement unit performs a predetermined clustering process based on the position information and the label stored in the storage unit, and places the label for each cluster.
The adjusting means optimizes the unity on the map of each cluster arranged by the arranging means within a range in which a predetermined accuracy of the contents of the cluster on the map can be ensured. The label placement apparatus according to claim 1, wherein the placement of the clusters is adjusted.

Images