A method, device, storage medium and computing device for identifying a news occurrence location

Abstract

The present invention provides a method, device, storage medium and computing device for identifying a news location. The method comprises: extracting multiple entity information from a news text; if any entity information is location information, constructing a first location tree according to the entity information; if any entity information is not location information, analyzing location information related to the entity information, constructing a second location tree according to the location information associated with the entity information; fusing the location information respectively contained in the first location tree and the second location tree, and identifying the location of the news based on the fused location information. The location information related to the entity information in the constructed second location tree can effectively supplement the location information in the news text, thereby improving the recognition accuracy and recall rate of the news location in the news text.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to a method, device, storage medium and computing equipment for identifying a news location. Background Art

[0002] Every day, a large amount of news texts are generated across the country and around the world. These news texts may have a data volume of millions. Such a huge amount of data will consume a lot of manpower and material resources if it relies solely on manual identification of the place where the news occurs. Therefore, it is necessary to use machines to automatically identify the place where the news occurs from the news texts. However, there are also various difficulties in the process of automatically identifying the place where the news occurs.

[0003] Usually, if you want to obtain the location where news occurs, the ideal situation is that the news only mentions one location, and it is the location where the news occurs. However, this is rarely the case in actual news. Often, in addition to the location where the news occurs, a piece of news also involves location information that is unrelated to the location where the news occurs. In this case, it is difficult to accurately determine the location of the news from multiple locations. In addition, there may be a part of the news text in the massive amount of news text that does not explicitly mention the location, and in this case, it is impossible to quickly and effectively obtain the location of the news. In addition, some news will also involve multiple news locations, such as the war between country A and country B. In this case, these news do not clearly define the location where the news occurs, which affects the identification process of the location where the news occurs. In the face of the above problems, there is an urgent need for a way to effectively determine the location where the news occurs to solve the above problems. Summary of the invention

[0004] In view of the above problems, the present invention is proposed to provide a method, apparatus, storage medium and computing device for identifying the place where news occurs that overcomes the above problems or at least partially solves the above problems. The method, apparatus, storage medium and computing device can use the second place tree to effectively supplement the place information in the news text, so that the place information content based on which the news place is identified is more comprehensive, thereby improving the recognition accuracy and recall rate of the news place in the news text.

[0005] According to one aspect of an embodiment of the present invention, a method for identifying a news location is provided, comprising:

[0006] Extract multiple entity information from news text;

[0007] If any of the entity information is location information, constructing a first location tree according to the entity information;

[0008] If any of the entity information is not location information, analyzing the location information associated with the entity information, and constructing a second location tree according to the location information associated with the entity information;

[0009] The location information respectively included in the first location tree and the second location tree is fused, and the location where the news occurred is identified based on the fused location information.

[0010] Optionally, constructing a first location tree according to the entity information includes:

[0011] Searching whether there is a standard place name identical to the name of the entity information in the preset place database, wherein the preset place database contains the standard place names corresponding to the place information;

[0012] If so, construct a first location tree using the entity information as the node name of the tree node;

[0013] If not, a standard place name corresponding to the entity information is searched from the preset place database, and the standard place name is used as a node name of a tree node to construct a first place tree.

[0014] Optionally, constructing a first location tree according to the entity information includes:

[0015] Counting the occurrence frequency of each location information used to construct the first location tree;

[0016] If any location information does not belong to the highest level, the occurrence frequency of the upper level location of the location information is increased by the same number of counting units as the occurrence frequency of the location information, and the counting is recursively repeated upwards in sequence until the highest level location information is counted and the counting is stopped;

[0017] The location information and the corresponding parent location information are used as tree nodes, and the node value is set for the corresponding tree node according to the frequency of occurrence of the location information;

[0018] The location information of the same level is located at the same tree level, and the location information of different levels is located at different tree levels.

[0019] Optionally, increasing the occurrence frequency of the upper level location of any location information by the same number of counting units as the occurrence frequency of any location information includes: if any location information has at least two upper level locations, increasing the occurrence frequencies of the at least two upper level locations by the same number of counting units as the occurrence frequency of any location information; taking each location information and the corresponding upper level location information as tree nodes includes: taking any location information and the at least two upper level location information as tree nodes; or

[0020] Increasing the occurrence frequency of the parent location of any location information by a counting unit equal to the number of occurrence frequencies of any location information includes: if any location information has at least two parent locations, extracting other entity information related to any location information from the news text, selecting one parent location related to the other entity information from at least two parent locations, and increasing the occurrence frequency of the selected parent location by a counting unit equal to the number of occurrence frequencies of any location information; taking each location information and the corresponding parent location information as tree nodes includes: taking any location information and the selected parent location information as tree nodes.

[0021] Optionally, after taking each location information and the corresponding upper-level location information as tree nodes and setting node values ​​for corresponding tree nodes according to the occurrence frequency of the location information, the method further includes:

[0022] According to the normalization formula Performing hierarchical normalization processing on the node values ​​contained in each tree level in the first location tree;

[0023] Among them, weightD represents the node value of the tree node D in the first location tree before normalization, SD1 represents the tree node set of the tree level to which the tree node D belongs, and D1 represents the node value of the tree node D after normalization.

[0024] Optionally, analyzing location information associated with the entity information includes:

[0025] Determine the text weight of each entity information in the news text according to a preset weight determination rule;

[0026] Ranking the text weights corresponding to each entity information, and determining the entity information whose text weight value is ranked before a preset ranking as the target entity;

[0027] The location information associated with the target entity is searched from a preset entity location library, wherein the preset entity location library includes a plurality of location information and entity information associated with each location information.

[0028] Optionally, fusing the location information respectively included in the first location tree and the second location tree includes:

[0029] Determine the first weight of each entity information in the first location tree in the news text according to the preset weight determination rule, and sum the first weights of each entity information to obtain the first tree weight of the first location tree;

[0030] Determine the second weight of the entity information associated with each location information in the second location tree in the news text according to the preset weight determination rule, and sum the second weights of each entity information to obtain the second tree weight of the second location tree;

[0031] Merge the tree nodes with the same node name in the first location tree and the second location tree, and retain the tree nodes with different node names, to obtain a fused location tree;

[0032] According to the formula Calculate the node weight of each tree node in the fused location tree, where α represents the first tree weight, β represents the second tree weight, P1 represents the normalized node value of the tree node P in the first location tree, P2 represents the normalized node value of the tree node P in the second location tree, SP1 represents the set of tree nodes of the tree level to which the tree node P in the first location tree belongs, and SP2 represents the set of tree nodes of the tree level to which the tree node P in the second location tree belongs.

[0033] Optionally, the preset weight determination rule includes:

[0034] Analyze the sentence components corresponding to any entity information in different sentences; determine the importance value of the entity information in the news text based on the preset component weights of each sentence component and the number of times the entity information appears in different sentence components; and / or

[0035] Analyze the positions of different sentences belonging to any entity information in the news text paragraph, and determine the importance value of the entity information in the news text based on the preset position weight of each sentence position and the number of times the sentence belonging to the entity information appears at different positions in the news text paragraph; and / or

[0036] Analyze the positions of different paragraphs to which any entity information belongs in the news text, and determine the importance value of the entity information in the news text based on the preset position weight of each paragraph position and the number of times the paragraph to which the entity information belongs appears in different positions of the news text.

[0037] Optionally, identifying the location where the news occurred based on the fused location information includes:

[0038] Selecting a tree node with the largest node weight in the first tree level of the fused location tree in a tree level order from top to bottom, wherein the first tree level corresponds to the highest level of location information, and the upper level node of the tree node in the first tree level is the root node;

[0039] Sum the node weights of the tree nodes in the next tree level of the branches contained in the tree node with the largest node weight in the first tree level;

[0040] If it is determined that the summed weight value reaches the preset weight value, continue to select the tree node with the largest node weight in the next tree level, and determine whether the sum of the node weights of the tree nodes in the next tree level of the branches contained in the tree node with the largest node weight in the next tree level reaches the preset weight value, until the summed weight value does not reach the preset weight value;

[0041] If it is determined that the summed weight value does not reach the preset weight value, the location information corresponding to the tree node with the largest node weight selected in the previous tree level of the next tree level is used as the place where the news occurred.

[0042] According to another aspect of an embodiment of the present invention, there is also provided a device for identifying a news location, comprising:

[0043] An extraction module, suitable for extracting multiple entity information from news text;

[0044] A first construction module, adapted to construct a first location tree according to any of the entity information if the entity information is location information;

[0045] a second construction module adapted to analyze the location information associated with the entity information if any of the entity information is not location information, and construct a second location tree according to the location information associated with the entity information;

[0046] The identification module is adapted to fuse the location information respectively contained in the first location tree and the second location tree, and identify the location where the news occurs based on the fused location information.

[0047] According to another aspect of an embodiment of the present invention, a computer storage medium is provided, wherein the computer storage medium stores a computer program code, and when the computer program code runs on a computing device, the computing device is caused to execute the method for identifying the location of news occurrence of any of the above embodiments.

[0048] According to another aspect of an embodiment of the present invention, a computing device is also provided, including: a processor; a memory storing computer program code; when the computer program code is executed by the processor, the computing device executes the news location identification method of any of the above embodiments.

[0049] After the embodiment of the present invention extracts multiple entity information from the news text, if any entity information is determined to be location information, a first location tree is constructed based on the entity information. If any entity information is determined not to be location information, the location information related to the entity information can be analyzed first, and then a second location tree can be constructed based on the location information associated with the entity information obtained by the analysis. Then, the location information respectively contained in the first location tree and the second location tree can be fused, and finally the location where the news occurred can be identified based on the fused location information. By constructing a location tree containing location information, the occurrence of news in the news text can be more accurately identified based on the tree node information in the location tree. Moreover, the location information related to the entity information in the constructed second location tree can effectively supplement the location information in the news text, so that the location information content based on which the news occurred is identified is more comprehensive, thereby improving the recognition accuracy and recall rate of the news location in the news text.

[0050] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below.

[0051] Based on the following detailed description of specific embodiments of the present invention in conjunction with the accompanying drawings, those skilled in the art will become more aware of the above and other objects, advantages and features of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0053] Figure 1 A schematic diagram showing a flow chart of a method for identifying a news location according to an embodiment of the present invention;

[0054] Figure 2 A schematic diagram of the structure of a location tree according to an embodiment of the present invention is shown;

[0055] Figure 3 A tree structure schematic diagram showing a process of normalizing node values ​​of tree nodes in a location tree according to an embodiment of the present invention is shown;

[0056] Figure 4 A schematic diagram showing the fusion of a first place tree and a second place tree according to an embodiment of the present invention is shown;

[0057] Figure 5 A schematic structural diagram of a device for identifying a news location according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0058] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

[0059] In order to solve the above technical problems, the embodiment of the present invention provides a method for identifying the location where news occurs. Figure 1 FIG. 2 is a flow chart showing a method for identifying a news location according to an embodiment of the present invention. Figure 1 The method for identifying the location where news occurs includes steps S102 to S108.

[0060] Step S102, extracting multiple entity information from the news text.

[0061] The entity information of the embodiment of the present invention may include entity information such as person information, other biological information, location information, organization information, etc. The preset location database includes multiple location information. The embodiment of the present invention can extract multiple entity information from the news text, and match each extracted entity information with the location information in the preset location database. If the location information consistent with any entity information is matched, it can be determined that the entity information is location information.

[0062] Step S104: if any entity information is location information, construct a first location tree according to the entity information.

[0063] Step S106: if any entity information is not location information, analyze the location information associated with the entity information, and construct a second location tree according to the location information associated with the entity information.

[0064] In this step, since the main body of the news often appears as an entity and the entity information may be associated with location information, the embodiment of the present invention can accurately identify the place where the news occurs by combining the location information in the news text and the location information associated with the entity information other than the location information.

[0065] The first location tree and the second location tree of the embodiment of the present invention represent location data (including location information associated with entity information and entity information that is location information itself) using a multi-branch tree data structure. Location information associated with entity information and entity information that is location information itself can all be used as tree nodes of the location tree, and each tree node forms a tree structure. In addition, each location tree can include a root node and multiple levels of child nodes (i.e., multiple levels of tree nodes), child nodes of the same level are located at the same tree level, and child nodes of different levels are located at different tree levels. The embodiment of the present invention can define the first tree level to correspond to the highest level of location information, and the levels of tree nodes corresponding to other tree levels decrease step by step, and the upper level node of the tree node in the first tree level is the root node. For example, the tree node in the first tree level corresponds to national-level location information, the tree node in the second tree level corresponds to provincial-level location information, the tree node in the third tree level corresponds to municipal-level location information, and so on. If a tree node is a child node, then the upper level node of the node is the parent node, each child node has one and only one parent node, and there is no loop in the location tree.

[0066] See also Figure 2 For example, a location tree contains a root node, and the location information corresponding to each level of tree nodes (each level of sub-nodes) is "United States", "China", "Zhejiang", "Guangdong", "Shanghai", and "Hangzhou", among which "United States" and "China" belong to the same level, "Zhejiang", "Guangdong", and "Shanghai" belong to the same level and are all sub-nodes of the next level of "China", and "Hangzhou" belongs to the next level of "Zhejiang".

[0067] Step S108, fusing the location information respectively included in the first location tree and the second location tree, and identifying the location where the news occurred based on the fused location information.

[0068] The embodiment of the present invention can more accurately identify the occurrence of news in a news text based on the tree node information in the location tree by constructing a location tree containing location information. Moreover, the location information related to the entity information in the constructed second location tree can effectively supplement the location information in the news text, making the location information content based on which the news occurrence location is identified more comprehensive, thereby improving the recognition accuracy and recall rate of the news occurrence location in the news text.

[0069] It should be noted that if the entity information extracted from the news text is not location information, that is, the location information is not explicitly mentioned in the news text, so that the first location tree cannot be effectively constructed, then in the process of fusing the location information respectively contained in the first location tree and the second location tree, the fusion result obtained is the location information contained in the second location tree, so as to identify the location where the news occurred based on the location information contained in the second location tree. Therefore, even if the location information is not directly mentioned in the news text, the location where the news occurred in the news text can be accurately and effectively identified by using the embodiment of the present invention.

[0070] Referring to step S102 above, in one embodiment of the present invention, when extracting multiple entity information from a news text, the NLP (Natural Language Processing) method can be used to extract the entity information in the news text. Of course, other methods can also be used to extract the entity information. The embodiment of the present invention does not specifically limit the method for extracting the entity information.

[0071] Sometimes, different names are used for the same place in news texts. For example, some news reports call Shanghai "Magic City", while others simply call it Shanghai. Another example is Chongqing, which is called "Hot Pot City" by some news reports, while others simply call it Chongqing. Another example is the United States, which is directly called the United States by some news reports, while others call it the United States of America. Therefore, in order to facilitate the construction of the first place tree and make the place information names in the constructed first place tree consistent, in an embodiment of the present invention, when executing step S104 to construct the first place tree, different names for the same place can be unified into the same standard place name, thereby constructing the first place tree based on the standard place name.

[0072] Specifically, first search the preset location database for a standard location name that is identical to the name of the extracted entity information. If the preset location database contains a standard location name that is identical to the name of the extracted entity information, it indicates that the name of the entity information extracted from the news text is the standard location name, and the entity information can be directly used as the node name of the tree node to construct the first location tree. If the preset location database does not contain a standard location name that is identical to the name of the extracted entity information, it indicates that the name of the entity information extracted from the news text is not a standard location name, and the standard location name corresponding to the entity information can be searched from the preset location database, and then the standard location name can be used as the node name of the tree node to construct the first location tree.

[0073] In this embodiment, the preset location database includes standard location names corresponding to location information. Since it has been determined that the extracted entity information is location information, the standard location name corresponding to the extracted entity information can be searched from the preset location database.

[0074] For example, the name of the extracted entity information is "imperial capital", and there is no such name in the standard place names in the preset place database. At this time, the standard place name corresponding to "imperial capital" is found in the preset place database as "Beijing", so the standard place name "Beijing" can be used as the node name of the tree node to construct the first place tree. The node names in the first place tree thus constructed are unified, which helps to quickly identify the place where the news occurred later.

[0075] Of course, in the construction process of the second location tree, the location information with the standard location name associated with the entity information can be directly analyzed, or after analyzing the location information associated with the entity information, the location information associated with the entity information can be unified into the standard location name in the manner of the above embodiment. The embodiment of the present invention does not make any specific limitation on this.

[0076] In an embodiment of the present invention, the specific process of using the entity information as a tree node and constructing the first location tree in step S104 above may include the following content.

[0077] First, the occurrence frequency of each location information used to construct the first location tree is counted.

[0078] Then, if it is analyzed that any location information does not belong to the highest level, the occurrence frequency of the upper level location of the location information can be increased by the same number of counting units as the occurrence frequency of any location information (such as one counting unit is 1 time), and the occurrence frequency of the lower level location of the location information remains unchanged, and the counting is recursively counted upwards in sequence until the counting of the location information of the highest level is completed and the counting is stopped. Counting recursively upwards in sequence until the counting of the location information of the highest level is completed and the counting is stopped means that, assuming that the occurrence frequency of a non-national-level location information is 2 times, after the occurrence frequency of the upper level location information of the location information is increased by 2 times, if it is continued to be determined that the upper level location information still has the upper level location information, then the occurrence frequency of the upper level location information will also be increased by 2 times, and so on. When the location information that has been increased by 2 times for the last time is the location information of the highest level, the recursive counting can be stopped.

[0079] Finally, each location information and the corresponding parent location information are used as tree nodes, and node values ​​are set for the corresponding tree nodes according to the frequency of occurrence of the location information, wherein location information of the same level is located at the same tree level, and location information of different levels is located at different tree levels.

[0080] In this embodiment, for example, the upper level location of Hangzhou is Zhejiang, the upper level location is China, and the lower level location of Hangzhou is Xihu District. Assuming that the actual place where the news content in the news text occurs is Hangzhou, and the news text mentions Zhejiang once, Guangdong twice, and Hangzhou 10 times, the highest level is defined as a provincial place. For the location information "Hangzhou" which is not the highest level, the frequency of occurrence of the upper level location "Zhejiang" of Hangzhou can be increased by 10 times. In the final statistical frequency of occurrence, Zhejiang appears 11 times, Hangzhou 10 times, and Guangdong 2 times. Since Zhejiang appears the most frequently, it can be ensured that the Zhejiang branch is entered to identify the news occurrence when the news occurrence is subsequently identified. If the frequency of Zhejiang's occurrence is not increased according to the frequency of Hangzhou's occurrence, then since Zhejiang and Guangdong are at the same level, Guangdong appears more frequently than Zhejiang, then the Guangdong branch will be entered when the news occurrence is subsequently identified, and the news occurrence cannot be accurately identified.

[0081] In an embodiment of the present invention, the principle of using the location information associated with the entity information as a tree node and constructing the second location tree is basically the same as the principle of constructing the first location tree described above. Specifically, first, the frequency of occurrence of each location information associated with the entity information used to construct the second location tree is counted. Then, if any location information does not belong to the highest level, the frequency of occurrence of the upper level location of any location information is increased by the same number of counting units as the number of occurrence frequencies of any location information, and the number is recursively counted to the upper level in sequence until the highest level location information is counted and the counting is stopped after the counting is completed. Then, each location information and the corresponding upper level location information are used as tree nodes, and the node value is set for the corresponding tree node according to the frequency of occurrence of the location information. For details, please refer to the above embodiment, and no specific limitation is made here.

[0082] Some places in the news text may have the same name. For example, the place information extracted from the news text is the city information of "S", but there is a city "S" in country A and there is also a city "S" in country B, that is, it is possible that the same place information has at least two upper-level places. In this regard, the embodiments of the present invention include two methods for processing cities with the same name.

[0083] Method 1: If any location information has at least two upper-level locations, the occurrence frequencies of the at least two upper-level locations can be increased by the same number of counting units as the occurrence frequency of the location information. In addition, any location information and the at least two upper-level location information are both used as tree nodes.

[0084] For example, the city "S" appears three times in the news text, but the news text does not clearly state whether it is a city in country B or country A. You can directly add the city "S", country A, and country B as tree nodes to the location tree, and set the frequency of occurrence of the city "S", country A, and country B to 3 times respectively. If the news text is actually for the city "S" in country A, it can still be identified as country A with a high probability based on the location information related to other entity information. This embodiment can increase the frequency of occurrence of location information in order from low-level location information to high-level location information, that is, first increase the frequency of occurrence of the city "S", and then increase the frequency of occurrence of country A and country B at the same time. This method can simply and quickly process cities with the same name.

[0085] Method 2: If any location information has at least two parent locations, directly extract other entity information related to the location information from the news text, select a parent location related to the other entity information from the at least two parent locations, and increase the occurrence frequency of the selected parent location by the same number of counting units as the occurrence frequency of the location information. In addition, any location information and the selected parent location information are used as tree nodes. That is, when there are locations with the same name, this method only puts one parent location information of the same name into the location tree.

[0086] For example, the city "S" appears three times in the news text, and other entity information related to the "S" extracted from the news text includes "S Cathedral" and "S Bridge", and the upper-level place related to the entity information "S Cathedral" and "S Bridge" is Country A. Therefore, the city "S" and the selected country A can be added to the place tree, and the frequency of occurrence of the city "S" and country A can be set to 3 times. Therefore, this method can first extract entity information related to any place information (including other place information) from the main body of the news text, and then make a correlation judgment on the entity information and the upper-level place information, so as to select the upper-level place information with a higher correlation with the entity information and put it into the place tree. This method can more accurately determine the upper-level place of the place with the same name.

[0087] In one embodiment of the present invention, after setting the node value for the corresponding tree node according to the frequency of occurrence of the location information, the normalization formula The node values ​​contained in each tree level in the first location tree are subjected to hierarchical normalization processing; wherein weightD represents the node value of the tree node D in the first location tree before normalization, SD1 represents the tree node set of the tree level to which the tree node D belongs, and D1 represents the node value of the tree node D after normalization. Moreover, the node value after normalization is between 0 and 1.

[0088] Of course, the node values ​​contained in each tree level in the second location tree will also be subjected to hierarchical normalization processing according to the normalization method in this embodiment, which will not be described in detail here.

[0089] See also Figure 3 For example, the node values ​​of each tree node in the location tree before normalization (such as the first location tree or the second location tree) are 1 for the United States, 4 for China, 1 for Beijing, and 2 for Shanghai. Among them, the count of the node value of China is: Beijing is mentioned 1 time, Shanghai is mentioned 2 times, and China is directly mentioned 1 time in the news text, and the sum of the values ​​(i.e. 1+2+1) is equal to 4 times. The node values ​​of each tree node in the location tree after normalization are 0.2 for the United States, 0.8 for China, 0.33 for Beijing, and 0.67 for Shanghai.

[0090] In combination with the above embodiments, since in the process of constructing the place tree, the frequency of occurrence of the tree nodes is counted in sequence toward the superior places, if the place tree is globally normalized, a total weight problem will arise, and the greater the depth of the node, the greater the impact of the total weight. Therefore, when the embodiment of the present invention normalizes the node values ​​of the tree nodes in the first place tree and the second place tree, hierarchical normalization is performed according to each tree level of the place tree, that is, the places at the same tree level (that is, the same depth) in the place tree are normalized, so that the tree nodes at each level in the place tree do not affect each other.

[0091] Referring to step S106 above, in one embodiment of the present invention, in the process of analyzing the location information associated with the entity information, the text weight of each entity information in the news text can be determined according to the preset weight determination rule. Then, the text weights corresponding to each entity information are ranked, and the entity information whose text weight value is ranked before the preset ranking is determined as the target entity. Then, the location information associated with the target entity is searched from the preset entity location library, wherein the preset entity location library contains multiple location information and entity information associated with each location information.

[0092] In this embodiment, the entity information whose text weight value ranking is before the preset ranking is the entity information with a higher degree of importance in the news text. By searching for the location information associated with the entity information with a higher degree of importance, the location information in the news text can be determined with a higher probability, thereby facilitating the subsequent more accurate identification of the place where the news occurred. For example, the preset ranking is the first, and the target entity with the first text weight value ranking is "Company A". By searching for the location information associated with "Company A" in the preset entity location library, it is known that Company A is located in Beijing, so "Beijing" can be used as a tree node of the second location tree.

[0093] In an optional embodiment, the preset weight determination rule may include the following rules.

[0094] The first method is to analyze the sentence components corresponding to any entity information in different sentences, and then determine the importance of the entity information in the news text based on the preset component weights of each sentence component and the number of times the entity information appears in different sentence components. Among them, sentence components are the components of a sentence, and sentence components usually include eight types: subject, predicate, object, verb, attributive, adverbial, complement and center. Generally, entity information that appears as a subject is relatively important, so the weight of the subject component can be set higher, and the weight of other components such as the object can be set relatively low. For example, the entity information "Expo" in the news text appears 3 times as a subject component and 2 times as a predicate component in different sentences. The component weight of the subject multiplied by 3 plus the component weight of the predicate multiplied by 2 can get the importance of "Expo" in the news text.

[0095] The second method is to analyze the positions of different sentences belonging to any entity information in the news text paragraphs, and then determine the importance of the entity information in the news text based on the preset position weight of each sentence position and the number of times the sentences belonging to the entity information appear in different positions of the news text paragraph. Usually, the sentences at the beginning of a paragraph are relatively important, so the position weight of the sentences at the beginning of a paragraph is higher, and the position weights of other positions are lower. For example, the positions of different sentences belonging to the entity information "Expo" in the news text are 2 sentences as the first sentences at the beginning of different paragraphs, and 1 sentence as the sentence in the middle of the paragraph. The position weight at the beginning of the paragraph multiplied by 2 plus the position weight in the middle of the paragraph multiplied by 1 can get the importance of "Expo" in the news text.

[0096] The third method is to analyze the positions of different paragraphs to which the entity information belongs in the news text for any entity information, and determine the importance of the entity information in the news text based on the preset position weight of each paragraph position and the number of times the paragraph to which the entity information belongs appears in different positions of the news text. Usually, the first paragraph of the news is relatively important, so the position weight of the first paragraph of the news is higher, and the position weights of other positions are lower. For example, the positions of different paragraphs to which the entity information "Exposition" belongs in the news text are: 2 "Exposition" paragraphs are the first paragraph of the text, and 1 "Exposition" paragraph is the middle paragraph of the text. The position weight of the first paragraph multiplied by 2 plus the position weight of the middle paragraph multiplied by 1 can get the importance of "Exposition" in the news text.

[0097] Of course, any at least two of the above rules may be combined to determine the text weight of the entity information in the news text.

[0098] In an embodiment of the present invention, after determining the importance value of the location information associated with each entity information in the news text, the text weight of each entity information in the news text in the above embodiment can be obtained by normalizing the determined importance value. After determining the importance value of the entity information in the first location tree in the news text, the first weight of the first location tree in the above embodiment can be obtained by normalizing the determined importance value. After determining the importance value of the entity information associated with each location information in the second location tree in the news text, the second weight of the second location tree in the above embodiment can be obtained by normalizing the determined importance value.

[0099] Referring to step S108 above, in one embodiment of the present invention, the process of fusing the location information respectively included in the first location tree and the second location tree is specifically as follows.

[0100] First, the first weight of each entity information in the first place tree in the news text is determined according to the preset weight determination rule, and the first weight of each entity information is summed to obtain the first tree weight of the first place tree. That is, the first weight of each entity information is weighted to obtain the first tree weight of the first place tree.

[0101] In this embodiment, each entity information in the first place tree includes all place information in the news text used to construct the first place tree, that is, the place information in the news text that has the same name as the tree node in the first place tree (that is, the standard place information in the news text), or the place information in the news text that corresponds to the name of the tree node in the first place tree (that is, the non-standard place information in the news text). The first weight of each entity information in the first place tree in the news text is determined according to the preset weight determination rule, that is, the first weight of all place information in the news text used to construct the first place tree is determined in the news text.

[0102] Then, the second weight of the entity information associated with each location information in the second location tree in the news text is determined according to the preset weight determination rule, and the second weight of each entity information is summed to obtain the second tree weight of the second location tree. That is, the second weight of each entity information is weighted to obtain the second tree weight of the second location tree.

[0103] In this embodiment, the second weight of the entity information associated with the standard place name in the second place tree in the news text may be determined first, and the second weight may be weighted to obtain the second tree weight of the second place tree.

[0104] The preset weight determination rule in this embodiment can refer to the relevant content introduced in the above embodiment, and will not be repeated here.

[0105] Furthermore, tree nodes with the same node name in the first place tree and the second place tree are merged, and tree nodes with different node names are retained to obtain a fused place tree.

[0106] See also Figure 4 The tree nodes "China" and "Beijing" in the first place tree also exist in the second tree node. The "China" and "Beijing" in the two place trees are merged. The first place tree also contains the tree node "Shanghai". The second place tree also contains the tree nodes "USA" and "Hangzhou". The fused place tree obtained after merging the two place trees contains "China", "Beijing", "Shanghai", "USA" and "Hangzhou".

[0107] Finally, based on the weights of the first and second trees calculated above, according to the formula Calculate the node weight of each tree node in the fused location tree. α represents the first tree weight, β represents the second tree weight, P1 represents the normalized node value of the tree node P in the first location tree, P2 represents the normalized node value of the tree node P in the second location tree, SP1 represents the tree node set of the tree level to which the tree node P in the first location tree belongs, and SP2 represents the tree node set of the tree level to which the tree node P in the second location tree belongs. In addition, the first tree weight is the tree weight calculated after the tree nodes of the first location tree are normalized, and the second tree weight is the tree weight calculated after the tree nodes of the second location tree are normalized.

[0108] For example, if the tree node P represents "Beijing", the node weight of the tree node "Beijing" in the fused location tree can be calculated according to the first tree weight and the second tree weight, and based on the normalized node value of the tree node "Beijing" in the first location tree and the normalized node value of the tree node "Beijing" in the second location tree according to the above formula.

[0109] In this embodiment, in the process of identifying the place where the news occurred based on the fused location information, the tree node with the largest node weight can be selected in the first tree level of the fused location tree in the tree level order from top to bottom, wherein the first tree level corresponds to the highest level of location information, and the upper level node of the tree node in the first tree level is the root node. Then, the node weights of the tree nodes in the next tree level of the branches contained in the tree node with the largest node weight in the first tree level are summed, and it is determined whether the summed weight value reaches the preset weight value. For example, the tree level with the highest location level is the national level tree level, and according to the tree level order from top to bottom, the next tree level is the provincial or municipal level tree level. This embodiment follows the tree level order from top to bottom, that is, the tree node weights are summed up from high-level locations to low-level locations.

[0110] If it is determined that the summed weight value reaches the preset weight value, continue to select the tree node with the largest node weight in the next tree level, and determine whether the sum of the node weights of the tree nodes in the next tree level of the branches contained in the tree node with the largest node weight in the next tree level reaches the preset weight value, until the summed weight value does not reach the preset weight value. If it is determined that the summed weight value does not reach the preset weight value, the location information corresponding to the tree node with the largest node weight selected in the previous tree level of the next tree level is used as the place where the news occurred.

[0111] For example, combined with Figure 4 As shown in the content, the first tree level is the national level tree level, the next tree level is the provincial or municipal level tree level (the second tree level), and the next level is the prefectural city level tree level (the third tree level). If the tree node with the largest node weight in the first tree level is "China", the node weights of the tree nodes in the municipal level tree level containing branches of "China" are summed, that is, the node weights of "Beijing", "Hangzhou" and "Shanghai" are summed. If the weight value after summation reaches the preset weight value, the tree node with the largest node weight is selected from "Beijing", "Hangzhou" and "Shanghai". If the tree node with the largest node weight selected is "Beijing", "Beijing" is used as the place where the news is identified. If the weight value after summation does not reach the preset weight value, stop judging the next tree level, and use "China" as the place where the news is identified. Therefore, the use of the embodiment of the present invention can effectively improve the recognition accuracy and recall rate of the place where the news occurs in the news text.

[0112] Based on the same inventive concept, an embodiment of the present invention further provides a device for identifying a news location. Figure 5 FIG. 2 shows a schematic diagram of a structure of a device for identifying a news location according to an embodiment of the present invention. Figure 5 The news occurrence location identification device includes an extraction module 510, a first construction module 520, a second construction module 530, and an identification module 540.

[0113] The extraction module 510 is adapted to extract multiple entity information from the news text.

[0114] The first construction module 520 is adapted to construct a first location tree according to the entity information if any entity information is location information.

[0115] The second construction module 530 is adapted to analyze the location information associated with the entity information if any entity information is not location information, and construct a second location tree according to the location information associated with the entity information.

[0116] The identification module 540 is adapted to fuse the location information respectively included in the first location tree and the second location tree, and identify the location where the news occurred based on the fused location information.

[0117] In one embodiment of the present invention, the first construction module 520 is also suitable for: searching whether there is a standard place name that is the same as the entity information name in the preset place database, wherein the preset place database contains the standard place name corresponding to the place information; if so, constructing the first place tree using the entity information as the node name of the tree node; if not, searching the preset place database for the standard place name corresponding to the entity information, and constructing the first place tree using the standard place name as the node name of the tree node.

[0118] In one embodiment of the present invention, the first construction module 520 is also suitable for: counting the frequency of occurrence of each location information used to construct the first location tree; if any location information does not belong to the highest level, increasing the frequency of occurrence of the upper level location of any location information by the same counting unit as the number of the frequency of occurrence of any location information, and recursively counting upwards in sequence until the counting of the highest level location information is completed and the counting is stopped; taking each location information and the corresponding upper level location information as a tree node, and setting a node value for the corresponding tree node according to the frequency of occurrence of the location information; wherein, location information of the same level is located at the same tree level, and location information of different levels is located at different tree levels.

[0119] In one embodiment of the present invention, the first construction module 520 is also suitable for: if any location information has at least two upper-level locations, the occurrence frequency of the at least two upper-level locations is increased by the same counting unit as the number of occurrence frequencies of any location information, and any location information and at least two upper-level location information are used as tree nodes; or if any location information has at least two upper-level locations, other entity information related to any location information is extracted from the news text, and an upper-level location related to the other entity information is selected from at least two upper-level locations, the occurrence frequency of the selected upper-level location is increased by the same counting unit as the number of occurrence frequencies of any location information, and each location information and the corresponding upper-level location information are used as tree nodes, including: taking any location information and the selected upper-level location information as tree nodes.

[0120] In one embodiment of the present invention, the news location identification device further includes a normalization module adapted to calculate the news location according to the normalization formula: The node values ​​contained in each tree level in the first location tree are hierarchically normalized; wherein weightD represents the node value of the tree node D in the first location tree before normalization, SD1 represents the set of tree nodes of the tree level to which the tree node D belongs, and D1 represents the node value of the tree node D after normalization.

[0121] In one embodiment of the present invention, the second construction module 530 is also suitable for determining the text weight of each entity information in the news text according to a preset weight determination rule; ranking the text weights corresponding to each entity information, and determining the entity information whose text weight value is ranked before a preset ranking as the target entity; searching for location information associated with the target entity from a preset entity location library, wherein the preset entity location library contains multiple location information and entity information associated with each location information.

[0122] In one embodiment of the present invention, the identification module 540 is further adapted to determine the first weight of each entity information in the first place tree in the news text according to a preset weight determination rule, and sum the first weights of each entity information to obtain the first tree weight of the first place tree; determine the second weight of the entity information associated with each place information in the second place tree in the news text according to the preset weight determination rule, and sum the second weights of each entity information to obtain the second tree weight of the second place tree; merge the tree nodes with the same node names in the first place tree and the second place tree, and retain the tree nodes with different node names to obtain a fused place tree; according to the formula Calculate the node weight of each tree node in the fused location tree, where α represents the first tree weight, β represents the second tree weight, P1 represents the normalized node value of the tree node P in the first location tree, P2 represents the normalized node value of the tree node P in the second location tree, SP1 represents the set of tree nodes of the tree level to which the tree node P in the first location tree belongs, and SP2 represents the set of tree nodes of the tree level to which the tree node P in the second location tree belongs.

[0123] In one embodiment of the present invention, the preset weight determination rule includes: analyzing the sentence components corresponding to any entity information in different sentences; determining the importance value of the entity information in the news text based on the preset component weights of each sentence component and the number of times the entity information corresponds to different sentence components; and / or analyzing the positions of different sentences belonging to any entity information in the news text paragraphs, and determining the importance value of the entity information in the news text based on the preset position weights of each sentence position and the number of times the sentences belonging to the entity information appear at different positions in the news text paragraphs; and / or analyzing the positions of different paragraphs belonging to any entity information in the news text, and determining the importance value of the entity information in the news text based on the preset position weights of each paragraph position and the number of times the paragraphs belonging to the entity information appear at different positions in the news text.

[0124] In one embodiment of the present invention, the identification module 540 is also suitable for selecting a tree node with the largest node weight in the first tree level of the fused location tree in a tree level order from top to bottom, wherein the first tree level corresponds to the highest level of location information, and the upper-level node of the tree node in the first tree level is the root node; summing the node weights of the tree nodes in the next tree level of the branches contained in the tree node with the largest node weight in the first tree level; if it is judged that the summed weight value reaches the preset weight value, continue to select the tree node with the largest node weight in the next tree level, and judge whether the sum of the node weights of each tree node in the next tree level of the branches contained in the tree node with the largest node weight in the next tree level reaches the preset weight value, until the summed weight value does not reach the preset weight value; if it is judged that the summed weight value does not reach the preset weight value, the location information corresponding to the tree node with the largest node weight selected in the upper tree level of the next tree level is used as the place where the news occurred.

[0125] Based on the same inventive concept, an embodiment of the present invention further provides a computer storage medium storing computer program code. When the computer program code runs on a computing device, the computing device executes the news location identification method of any of the above embodiments.

[0126] Based on the same inventive concept, an embodiment of the present invention also provides a computing device, including: a processor; a memory storing computer program code; when the computer program code is executed by the processor, the computing device executes the news location identification method of any of the above embodiments.

[0127] Those skilled in the art can clearly understand that the specific working processes of the systems, devices, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and for the sake of brevity, they are not further described here.

[0128] In addition, the functional units in various embodiments of the present invention may be physically independent of each other, or two or more functional units may be integrated together, or all functional units may be integrated into one processing unit. The above integrated functional units may be implemented in the form of hardware, or in the form of software or firmware.

[0129] Those skilled in the art can understand that if the integrated functional unit is implemented in the form of software and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can essentially or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, which includes a number of instructions to enable a computing device (such as a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention when running the instructions. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk and other media that can store program codes.

[0130] Alternatively, all or part of the steps of implementing the aforementioned method embodiments may be accomplished by hardware associated with program instructions (such as a computing device such as a personal computer, a server, or a network device), and the program instructions may be stored in a computer-readable storage medium. When the program instructions are executed by a processor of a computing device, the computing device executes all or part of the steps of the methods described in the embodiments of the present invention.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that within the spirit and principles of the present invention, the technical solutions described in the aforementioned embodiments can still be modified, or some or all of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate from the protection scope of the present invention.

Claims

1. A method for identifying a news location, characterized in that: include: Extract multiple entity information from news text; If any of the entity information is location information, constructing a first location tree according to the entity information; If any of the entity information is not location information, analyzing the location information associated with the entity information, and constructing a second location tree according to the location information associated with the entity information; The location information respectively included in the first location tree and the second location tree is integrated, and the location where the news occurred is identified based on the integrated location information, including: Determine the first weight of each entity information in the first location tree in the news text according to the preset weight determination rule, and sum the first weight to obtain the first tree weight; determine the second weight of the entity information associated with each location information in the second location tree in the news text according to the preset weight determination rule, and sum the second weight to obtain the second tree weight; merge the tree nodes with the same node name in the first location tree and the second location tree, and retain the tree nodes with different node names to obtain a fused location tree, and according to the formula Calculate the node weight of each tree node in the fused location tree, where α represents the first tree weight, β represents the second tree weight, P1 represents the normalized node value of the tree node P in the first location tree, P2 represents the normalized node value of the tree node P in the second location tree, SP1 represents the tree node set of the tree level to which the tree node P in the first location tree belongs, and SP2 represents the tree node set of the tree level to which the tree node P in the second location tree belongs; In the order of tree levels from top to bottom, select a tree node with the largest node weight in the first tree level of the fused location tree, wherein the first tree level corresponds to the highest level of location information, and the upper level node of the tree node in the first tree level is the root node; sum the node weights of the tree nodes in the next tree level of the branches contained in the tree node with the largest node weight in the first tree level; If it is determined that the summed weight value reaches the preset weight value, continue to select the tree node with the largest node weight in the next tree level, and determine whether the sum of the node weights of the tree nodes in the next tree level of the branches contained in the tree node with the largest node weight in the next tree level reaches the preset weight value, until the summed weight value does not reach the preset weight value; If it is determined that the summed weight value does not reach the preset weight value, the location information corresponding to the tree node with the largest node weight selected in the previous tree level of the next tree level is used as the place where the news occurred.

2. The method according to claim 1, characterized in that: Constructing a first location tree according to the entity information includes: Searching whether there is a standard place name identical to the name of the entity information in a preset place database, wherein the preset place database contains standard place names corresponding to the place information; If so, construct a first location tree using the entity information as the node name of the tree node; If not, a standard place name corresponding to the entity information is searched from the preset place database, and the standard place name is used as a node name of a tree node to construct a first place tree.

3. The method according to claim 1, characterized in that The step of constructing a first location tree according to the entity information includes: Counting the occurrence frequency of each location information used to construct the first location tree; If any location information does not belong to the highest level, the occurrence frequency of the upper level location of the location information is increased by the same number of counting units as the occurrence frequency of the location information, and the counting is recursively repeated upwards in sequence until the highest level location information is counted and the counting is stopped; The location information and the corresponding parent location information are used as tree nodes, and the node value is set for the corresponding tree node according to the frequency of occurrence of the location information; The location information of the same level is located at the same tree level, and the location information of different levels is located at different tree levels.

4. The method according to claim 3, characterized in that Increasing the occurrence frequency of the upper level location of any location information by the same number of counting units as the occurrence frequency of any location information includes: if any location information has at least two upper level locations, increasing the occurrence frequencies of the at least two upper level locations by the same number of counting units as the occurrence frequency of any location information; taking each location information and the corresponding upper level location information as tree nodes includes: taking any location information and the at least two upper level location information as tree nodes; or Increasing the occurrence frequency of the parent location information of any location information by the same counting unit as the occurrence frequency of any location information includes: if any location information has at least two parent locations, extracting other entity information related to any location information from the news text, selecting one parent location related to the other entity information from at least two parent locations, and increasing the occurrence frequency of the selected parent location by the same counting unit as the occurrence frequency of any location information; taking each location information and the corresponding parent location information as tree nodes includes: taking any location information and the selected parent location information as tree nodes.

5. The method according to claim 3, characterized in that: After taking each location information and the corresponding parent location information as tree nodes and setting node values ​​for the corresponding tree nodes according to the occurrence frequency of the location information, the method further includes: According to the normalization formula Performing hierarchical normalization processing on the node values ​​contained in each tree level in the first location tree; Among them, weightD represents the node value of the tree node D in the first location tree before normalization, SD1 represents the tree node set of the tree level to which the tree node D belongs, and D1 represents the node value of the tree node D after normalization.

6. The method according to claim 5, characterized in that Analyzing location information associated with the entity information, including: Determine the text weight of each entity information in the news text according to a preset weight determination rule; Ranking the text weights corresponding to each entity information, and determining the entity information whose text weight value is ranked before a preset ranking as the target entity; The location information associated with the target entity is searched from a preset entity location library, wherein the preset entity location library includes a plurality of location information and entity information associated with each location information.

7. The method according to claim 6, characterized in that The preset weight determination rule includes: Analyze the sentence components corresponding to any entity information in different sentences; determine the importance value of the entity information in the news text based on the preset component weights of each sentence component and the number of times the entity information appears in different sentence components; and / or Analyze the positions of different sentences belonging to any entity information in the news text paragraph, and determine the importance value of the entity information in the news text based on the preset position weight of each sentence position and the number of times the sentence belonging to the entity information appears at different positions in the news text paragraph; and / or Analyze the positions of different paragraphs to which any entity information belongs in the news text, and determine the importance value of the entity information in the news text based on the preset position weight of each paragraph position and the number of times the paragraph to which the entity information belongs appears in different positions of the news text.

8. A device for identifying a news location, characterized in that: include: An extraction module, suitable for extracting multiple entity information from news text; A first construction module, adapted to construct a first location tree according to any of the entity information if the entity information is location information; a second construction module adapted to analyze the location information associated with the entity information if any of the entity information is not location information, and construct a second location tree according to the location information associated with the entity information; The identification module is suitable for fusing the location information respectively contained in the first location tree and the second location tree, and identifying the location where the news occurred based on the fused location information; wherein, the first weight of each entity information in the first location tree in the news text is determined according to a preset weight determination rule, and the first weight is summed to obtain the first tree weight; the second weight of the entity information associated with each location information in the second location tree in the news text is determined according to the preset weight determination rule, and the second weight is summed to obtain the second tree weight; the tree nodes with the same node names in the first location tree and the second location tree are merged, and the tree nodes with different node names are retained to obtain a fused location tree, and according to the formula Calculate the node weight of each tree node in the fused location tree, where α represents the first tree weight, β represents the second tree weight, P1 represents the normalized node value of the tree node P in the first location tree, P2 represents the normalized node value of the tree node P in the second location tree, SP1 represents the set of tree nodes of the tree level to which the tree node P in the first location tree belongs, and SP2 represents the set of tree nodes of the tree level to which the tree node P in the second location tree belongs; select the tree node with the largest node weight in the first tree level of the fused location tree in a tree level order from top to bottom, where the first tree level corresponds to the highest level of location information, and the upper-level node of the tree node in the first tree level is the root node ; Sum the node weights of the tree nodes in the next tree level of the branch contained in the tree node with the largest node weight in the first tree level; if it is judged that the summed weight value reaches the preset weight value, continue to select the tree node with the largest node weight in the next tree level, and judge whether the sum of the node weights of each tree node in the next tree level of the branch contained in the tree node with the largest node weight in the next tree level reaches the preset weight value, until the summed weight value does not reach the preset weight value; if it is judged that the summed weight value does not reach the preset weight value, the location information corresponding to the tree node with the largest node weight selected in the previous tree level of the next tree level is used as the place where the news occurred.

9. A computer storage medium storing a computer program code, which, when executed on a computing device, causes the computing device to execute the method for identifying the location of news as described in any one of claims 1 to 7.

10. A computing device comprising: processor; a memory storing computer program code; When the computer program code is executed by the processor, it causes the computing device to execute the method for identifying the place where news occurs according to any one of claims 1 to 7.

Images