News article origination identification method, apparatus and device

Abstract

The disclosure provides a news article first publication identification method, device and equipment, the method comprises the following steps: obtaining a news article to be analyzed, and extracting a title of the news article to be analyzed; according to the title, similar articles are screened from a news article set; when the similar articles are screened from the news article set, entity words in the news article to be analyzed and entity words in the similar articles are extracted respectively; the entity words in the news article to be analyzed and the entity words in the similar articles are matched, and whether the similar articles and the news article to be analyzed are articles of the same event is determined according to a matching result; when the similar articles and the news article to be analyzed are articles of the same event, a first publication article is determined from the articles of the same event as the news article to be analyzed, so that the accuracy of the news article first publication identification result can be improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of computer, and particularly relates to a news article first publication identification method, device and equipment. BACKGROUND

[0002] In a news media organization, the timeliness of a news report is very important, and a first publication of news usually has a great influence. Therefore, after a news article is published, the first publication of the news article is identified to predict the influence degree of the news article.

[0003] At present, the method for identifying the first publication of news includes a conventional search plus manual sorting method and a clustering method. In the method for identifying the first publication of news by using the conventional search plus manual sorting method, a search engine is used to search according to the keywords of a news article in a specified range, and then a manual judgment and screening are performed on the search results, the publication times of the news articles of the same event are compared and sorted, and thus the first publication of news at the analysis time and the specific timeliness are obtained. In the method for identifying the first publication of news by using the clustering method, the articles to be analyzed are clustered, and then the same news events are clustered in one class according to the clustering results, and thus the identification of the first publication of news and the timeliness analysis are performed in each class. However, the accuracy of the first publication identification result obtained by using the above method is poor. SUMMARY

[0004] Therefore, the present disclosure provides a news article first publication identification method, device and equipment, which can improve the accuracy of news article first publication identification.

[0005] According to a first aspect of the present disclosure, a news article first publication identification method is provided for identifying the first publication of an event in a news article, and the method comprises the following steps.

[0006] A news article to be analyzed is obtained, and a title of the news article to be analyzed is extracted;

[0007] According to the title, similar articles are screened from a news article set;

[0008] When the similar articles are screened from the news article set, entity words in the news article to be analyzed and entity words in the similar articles are extracted respectively;

[0009] The entity words in the news article to be analyzed and the entity words in the similar articles are matched, and whether the similar articles and the news article to be analyzed are articles of the same event is determined according to the matching result;

[0010] When the similar articles and the news article to be analyzed are articles of the same event, a first publication article is determined from the articles of the same event as the news article to be analyzed.

[0011] In a possible implementation, when the similar article is not filtered out from the set of news articles, the news article to be analyzed is directly determined as a first article.

[0012] In a possible implementation, when the similar article is filtered out from the set of news articles, the method further includes:

[0013] calculating and determining the similarity between the news article to be analyzed and the similar article;

[0014] When the similarity between the news article to be analyzed and the similar article is determined to exceed a preset threshold, the operation of extracting the entity words in the news article to be analyzed and the entity words in the similar article is performed again.

[0015] In a possible implementation, when the similarity between the news article to be analyzed and the similar article is determined to not exceed the preset threshold, the news article to be analyzed is determined as a first article.

[0016] In a possible implementation, when the similarity between the news article to be analyzed and the similar article is calculated, the calculation is performed based on a word vector model.

[0017] In a possible implementation, when the entity words in the news article to be analyzed and the entity words in the similar article are extracted, the extraction is implemented based on a sequence labeling model.

[0018] In a possible implementation, when the entity words in the news article to be analyzed and the entity words in the similar article are matched, and whether the similar article and the news article to be analyzed are articles of the same event are determined according to the matching result, the method includes:

[0019] extracting the same type of entity words from the entity words in the news article to be analyzed and the entity words in the similar article, and performing consistency matching;

[0020] When there are the same type of entity words and there are consistent entity words in the same type of entity words, it is determined that the similar article and the news article to be analyzed are articles of the same event.

[0021] In a possible implementation, when each of the similar articles is not an article of the same event as the news article to be analyzed, the news article to be analyzed is determined as a first article.

[0022] According to a second aspect of the present disclosure, a news article first article determination apparatus is provided for determining a first article of an event in a news article, and the apparatus includes:

[0023] The news article to be analyzed acquisition module is configured to acquire a news article to be analyzed and extract a title of the news article to be analyzed.

[0024] The similar article screening module is configured to screen similar articles from a news article set according to the title.

[0025] The entity word extraction module is configured to extract entity words in the news article to be analyzed and entity words in the similar articles, respectively, when the similar articles are screened from the news article set.

[0026] The same event article identification module is configured to match the entity words in the news article to be analyzed with the entity words in the similar articles and identify whether the similar articles and the news article to be analyzed are articles of the same event according to a matching result.

[0027] The first article identification module is configured to identify a first article from the articles of the same event as the news article to be analyzed when the similar articles and the news article to be analyzed are articles of the same event.

[0028] According to a third aspect of the present disclosure, a news article first article identification device is provided for identifying a first article of an event in a news article. The device comprises a processor, a memory for storing processor-executable instructions, and the processor is configured to execute the method according to the first aspect of the present disclosure.

[0029] In the present disclosure, a news article to be analyzed is acquired, a title of the news article to be analyzed is extracted, similar articles are screened from a news article set according to the title, entity words in the news article to be analyzed and entity words in the similar articles are extracted, respectively, when the similar articles are screened from the news article set, the entity words in the news article to be analyzed are matched with the entity words in the similar articles, and whether the similar articles and the news article to be analyzed are articles of the same event is identified according to a matching result, and a first article is identified from the articles of the same event as the news article to be analyzed when the similar articles and the news article to be analyzed are articles of the same event. In this way, the accuracy of the result of identifying a first article of a news article can be improved.

[0030] Other features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0031] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the present disclosure and serve to explain the principles of the present disclosure.

[0032] Figure 1A schematic flow chart of a news article first publication determining method according to an embodiment of the present disclosure is shown.

[0033] Figure 2 A schematic block diagram of a news article first publication determining apparatus according to an embodiment of the present disclosure is shown.

[0034] Figure 3 A schematic block diagram of a news article first publication determining device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0035] Various exemplary embodiments, features and aspects of the present disclosure will be explained in detail below with reference to the accompanying drawings. The same reference numbers in different drawings denote the same or similar elements. Although various aspects of the embodiments are illustrated in the drawings, the drawings are not necessarily drawn to scale unless specifically noted.

[0036] The word "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any implementation described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other implementations.

[0037] In addition, for the purpose of convenience and brevity, detailed descriptions of well-known functions and structures incorporated in the present disclosure can be omitted. It will be appreciated that the present disclosure can be practiced with the exact details as

[0038] <Method Embodiment>

[0039] Figure 1 A schematic flow chart of a news article first publication determining method according to an embodiment of the present disclosure is shown. As shown in Figure 1 the method comprises steps S1100-S1500.

[0040] S1100, a news article to be analyzed is acquired, and a title of the news article to be analyzed is extracted. The news article to be analyzed is a news article about an event to be analyzed.

[0041] S1200, according to the title of the news article to be analyzed, a similar article is screened from a news article set.

[0042] The news article set is a collection of published news articles. The news article set can store all published news articles, or can only store news articles published within a set time period, which is not specifically limited herein.

[0043] In the news article, the title of the news uses the most concise words to prompt the most valuable content in the news to the readers, which is the condensation and summary of the most valuable and important facts and issues in the news. If the titles of two news articles are similar, it means that the facts and issues reported in the two news articles are similar, so the news articles with similar titles can be used as similar articles of the news article to be analyzed.

[0044] In a possible implementation, the similar articles of the news article to be analyzed can be filtered from the news article set based on a search engine according to the title of the news article to be analyzed. The search engine can be a full-text search engine (i.e., an ES search engine), or other engines that can realize title similarity retrieval, which is not limited here.

[0045] In a possible implementation, when the similar articles of the news article to be analyzed are filtered from the news article set by the search engine, steps S1210-S1220 are included.

[0046] S1210, the search engine obtains retrieval information.

[0047] The retrieval information can include at least one of the title of the news article to be analyzed, retrieval conditions, and the number of returns. The retrieval conditions are the retrieval conditions of the similar articles, which can be retrieval statements representing title similarity. The number of returns is the number of similar articles returned by the search engine that meet the retrieval conditions. The number of returns can be set according to the specific application scenario. Preferably, the value range of the number of returns can be set to 10-30. Specifically, it can be set to 10, 20, or 30, which is not limited here.

[0048] S1220, filtering of similar articles according to the obtained retrieval information.

[0049] When there is no news article in the news article set that meets the retrieval conditions, the similar articles returned by the search engine are empty, that is, no similar articles of the news article to be analyzed are filtered from the news article set. In this case, the news article to be analyzed can be directly identified as the first article of the event to be analyzed.

[0050] When there is a news article in the news article set that meets the retrieval conditions, it indicates that there is a similar article of the current news article to be analyzed in the news article set. Therefore, the search engine will return similar articles according to the number of news articles that meet the retrieval conditions and the number of returns.

[0051] When the number of news articles meeting the search condition is less than the return number, the news articles meeting the search condition are directly extracted as similar articles. For example, when the number of news articles meeting the search condition is 20 and the return number is 30, the 20 news articles meeting the search condition are directly extracted as similar articles.

[0052] When the number of news articles meeting the search condition is greater than or equal to the return number, part of the news articles meeting the search condition are extracted as similar articles, and the number of returned similar articles is equal to the return number. For example, when the number of news articles meeting the search condition is 20 and the return number is 10, the 10 news articles with the highest similarity to the title of the news article to be analyzed are extracted from the news articles meeting the search condition as similar articles.

[0053] In a possible implementation, the similar articles are returned in the form of an article list. The article list can include one similar article or multiple similar articles, which is not limited herein. When the article list includes multiple similar articles, the multiple similar articles in the article list are arranged in descending order of similarity to the title of the news article to be analyzed.

[0054] In the embodiment in which the number of returned similar articles is equal to the return number, some similar articles can not be returned, which will result in omission of similar articles and affect the accuracy of the news article first publication identification.

[0055] In a possible implementation, to ensure the accuracy of the news article first publication identification, when the number of returned similar articles is equal to the return number, after the similar articles similar to the news article to be analyzed are screened from the news article set, the screened similar articles are updated and calibrated. That is, the screened similar articles are updated and calibrated to ensure that the similar articles screened from the news article set are more comprehensive, thereby avoiding inaccurate identification results due to omission of similar articles.

[0056] In a possible implementation, the screened similar articles can be updated and calibrated according to the size relationship between the minimum value of the similarity of the similar articles to the title of the news article to be analyzed and a preset calibration threshold. That is, when the minimum value of the similarity of the similar articles to the title of the news article to be analyzed is less than the calibration threshold, the screened similar articles do not need to be supplemented and updated. When the minimum value of the similarity of the similar articles to the title of the news article to be analyzed is greater than or equal to the calibration threshold, the screened similar articles need to be supplemented and updated. The calibration threshold can be set according to a specific application scenario. Preferably, the calibration threshold can be in the range of 0.7 to 0.8.

[0057] In the embodiment of including multiple similar news releases in the release list, when updating and calibrating the screened similar news releases, the following step is included: judging whether the similarity between the title of the last news release in the release list and the title of the news release to be analyzed is less than the calibration threshold:

[0058] In the case of being less than the calibration threshold, no supplementary update is needed for the screened similar news releases.

[0059] In the case of being greater than or equal to the calibration threshold, supplementary update is needed for the screened similar news releases. Specifically, the step of supplementary updating the screened similar news releases can include: first adjusting the return quantity in the search information, and then performing steps S1210-S1220 based on the adjusted return quantity until the number of returned similar news releases is less than the adjusted return quantity or the similarity between the title of the last news release in the returned release list and the title of the news release to be analyzed is less than the calibration threshold, and the news releases in the finally returned release list are taken as the similar news releases of the calibrated news release to be analyzed. The return quantity can be adjusted in a way of increasing by a multiple. For example, when the return quantity is configured as N in the nth search, the return quantity can be configured as 2*N in the (n+1)th search. Wherein, N and n are both integers greater than zero.

[0060] In a possible implementation, when the similar news releases are screened from the news release set, steps S1230-S1240 are further included.

[0061] S1230, calculating and judging the similarity between the news release to be analyzed and the similar news release.

[0062] In a possible implementation, the similarity between the news release to be analyzed and the similar news release can be calculated based on a word vector model. Specifically, the calculation step can include: inputting the title of the news release to be analyzed into the word vector model to obtain the word vector corresponding to the title of the news release to be analyzed. Inputting the title of the similar news release into the word vector model to obtain the word vector corresponding to the title of the similar news release. Calculating the similarity between the word vector corresponding to the title of the news release to be analyzed and the word vector corresponding to the title of the similar news release as the similarity between the news release to be analyzed and the similar news release.

[0063] After obtaining the similarity between the news release to be analyzed and the similar news release, it is judged whether the similarity between the news release to be analyzed and the similar news release is greater than a preset threshold. Wherein, the value of the preset threshold can be set according to the specific application scenario. Preferably, the value range of the calibration threshold can be 0.7-0.8.

[0064] In the embodiment of one similar news release, if it is judged that the similarity between the news release to be analyzed and the similar news release does not exceed the preset threshold, the news release to be analyzed can be determined as the first news release.

[0065] In the embodiment in which the number of similar articles is multiple, the similarity between the to-be-analyzed news article and each similar article needs to be calculated respectively in the reference step S1230, and it is determined whether the similarity with each similar article exceeds the preset threshold. When the similarity with each similar article does not exceed the preset threshold, it can be determined that the to-be-analyzed news article is a first article.

[0066] S1240, when it is determined that the similarity between the to-be-analyzed news article and the similar article exceeds the preset threshold, the operation of extracting the entity words in the to-be-analyzed news article and the entity words in the similar article is further performed.

[0067] In the embodiment in which the number of similar articles is multiple, when the similarity between the to-be-analyzed news article and some similar articles exceeds the preset threshold, the similar articles with the similarity exceeding the preset threshold are extracted from the to-be-analyzed news article, and the entity words in each extracted similar article and the entity words in the to-be-analyzed news article are extracted.

[0068] In the embodiment, the most similar similar article is further extracted from the screened similar articles by the preset threshold from the to-be-analyzed news article, so that the first article is determined in the most similar similar article through the steps of entity word extraction and matching. In this way, while ensuring the accuracy of the first article determination, the calculation amount of the entity word extraction and matching can be reduced, and the efficiency of the first article determination of the news article can be improved.

[0069] S1300, when the similar articles are screened from the news article set, the entity words in the to-be-analyzed news article and the entity words in the similar articles are extracted respectively.

[0070] In a possible implementation manner, the entity words in the to-be-analyzed news article and the entity words in the similar articles can be extracted by a sequence labeling model. The specific steps include: inputting the to-be-analyzed news article into the sequence labeling model to obtain the entity words in the to-be-analyzed news article and the types of the entity words. The similar articles are input into the sequence labeling model to obtain the entity words in the similar articles and the types of the entity words. The types of the entity words can include at least one of a person name, a place name, an organization name, and a time word.

[0071] In a possible implementation manner, after the entity words are extracted, a step of standardizing the entity words is further included. For example, the extracted place names are removed from the words such as “province”, “city”, “district”, and only the place names themselves are reserved. For another example, the extracted time words are unified into a unified time format. In this way, the accuracy of the subsequent entity word matching can be improved, and the accuracy of the first article determination of the news article can be improved.

[0072] S1400, matching the entity words in the news article to be analyzed with the entity words in the similar articles, and determining whether the similar articles and the news article to be analyzed are articles of the same event according to the matching result.

[0073] In a possible implementation, S1400 specifically includes the following steps: extracting entity words of the same type from the entity words in the news article to be analyzed and the entity words in the similar articles, and performing consistency matching.

[0074] When there are entity words of the same type and there are consistent entity words in the entity words of the same type, it is determined that the similar articles and the news article to be analyzed are articles of the same event. For example, the entity words in the news article to be analyzed include a person name A, and the similar articles also include the person name A, and thus the similar articles and the news article to be analyzed are articles of the same event.

[0075] When there are no entity words of the same type or there are no consistent entity words in the entity words of the same type, the similar articles and the news article to be analyzed are news articles of different events. For example, the entity words in the news article to be analyzed include a person name A and a place name B, and the similar articles do not include entity words of the person name and the place name, and thus the similar articles and the news article to be analyzed are news articles of different events. For another example, the entity words in the news article to be analyzed include a person name A, and the similar articles do not include the person name A, and thus the similar articles and the news article to be analyzed are news articles of different events.

[0076] In the embodiment in which there is one similar article, if the similar article and the news article to be analyzed are news articles of different events, it can be determined that the news article to be analyzed is a first-published article. In the embodiment in which there are multiple similar articles, when each of the similar articles is not an article of the same event as the news article to be analyzed, it can be determined that the news article to be analyzed is a first-published article.

[0077] S1500, when the similar articles and the news article to be analyzed are articles of the same event, determining a first-published article from the articles of the same event as the news article to be analyzed.

[0078] In a possible implementation, the news article with the earliest publishing time can be selected from the articles of the same event as the news article to be analyzed as the first-published article of the event to be analyzed.

[0079] In the present disclosure, a news article to be analyzed is acquired, and a title of the news article to be analyzed is extracted. Similar articles are screened from a news article set according to the title of the news article to be analyzed. Entity words in the news article to be analyzed and entity words in the similar articles are extracted respectively when the similar articles are screened from the news article set. The entity words in the news article to be analyzed and the entity words in the similar articles are matched, and it is determined whether the similar articles and the news article to be analyzed are articles of the same event according to a matching result. When the similar articles and the news article to be analyzed are articles of the same event, a first article is determined from the articles of the same event as the news article to be analyzed. In this way, the accuracy of a first article determination result of a news article can be improved.

[0080] <Device Embodiment>

[0081] Figure 2 A schematic block diagram of a news article first article determination device according to an embodiment of the present disclosure is shown.

[0082] As shown in Figure 2 The news article first article determination device 100 includes:

[0083] A news article to be analyzed acquisition module 110 is configured to acquire a news article to be analyzed, and extract a title of the news article to be analyzed.

[0084] A similar article screening module 120 is configured to screen similar articles from a news article set according to the title.

[0085] An entity word extraction module 130 is configured to extract entity words in the news article to be analyzed and entity words in the similar articles respectively when the similar articles are screened from the news article set.

[0086] A same event article determination module 140 is configured to match the entity words in the news article to be analyzed and the entity words in the similar articles, and determine whether the similar articles and the news article to be analyzed are articles of the same event according to a matching result.

[0087] A first article determination module 150 is configured to determine a first article from the articles of the same event as the news article to be analyzed when the similar articles and the news article to be analyzed are articles of the same event.

[0088] In a possible implementation, the similar article screening module 120 directly determines the news article to be analyzed as a first article when no similar article is screened from the news article set.

[0089] In a possible implementation, when the similar articles are screened from the news article set, the similar article screening module 120 is further configured to:

[0090] Calculate and determine the similarity between the news article to be analyzed and similar articles;

[0091] When it is determined that the similarity between the news article to be analyzed and similar articles exceeds a preset threshold, the operation of extracting entity words from the news article to be analyzed and entity words from similar articles is then performed.

[0092] In one possible implementation, the similar article screening module 120 determines that the news article to be analyzed is a first-release article when it determines that the similarity between the news article to be analyzed and similar articles does not exceed a preset threshold.

[0093] In one possible implementation, the similarity screening module 120 calculates the similarity between the news article to be analyzed and similar articles based on a word vector model.

[0094] In one possible implementation, the entity word extraction module 130 extracts entity words from the news article to be analyzed and from similar articles based on a sequence labeling model.

[0095] In one possible implementation, when the same event article identification module 140 matches entity words in the news article to be analyzed with entity words in similar articles, and confirms whether similar articles and the news article to be analyzed belong to the same event based on the matching results, it is specifically used for:

[0096] Extract similar entity words from the entity words of the news article to be analyzed and similar articles, and perform consistency matching;

[0097] If there are identical entity words of the same type, then the similar article and the news article to be analyzed are confirmed to be articles about the same event.

[0098] In one possible implementation, the first-release article identification module 150 identifies the news article to be analyzed as a first-release article when it determines that none of the similar articles belong to the same event as the news article to be analyzed.

[0099] <Equipment Example>

[0100] Figure 3 A schematic block diagram of a press release first-release certification device according to an embodiment of this disclosure is shown. Figure 3 As shown, the press release first publication determination device 200 includes a processor 210 and a memory 220 for storing executable instructions of the processor 210. The processor 210 is configured to implement any of the aforementioned press release first publication determination methods when executing the executable instructions.

[0101] It should be noted here that the number of processors 210 can be one or more. Meanwhile, the press release first publication device 200 in the embodiments of the present disclosure can also include an input device 230 and an output device 240. Among them, the processor 210, the memory 220, the input device 230 and the output device 240 can be connected through a bus, or can be connected through other ways, which is not specifically limited here.

[0102] The memory 220 as a kind of computer readable storage medium, it can be used to store software programs, computer executable programs and various modules, such as the program or module corresponding to the press release first publication method of the embodiments of the present disclosure. The processor 210 executes the software programs or modules stored in the memory 220, thereby performing various functional applications and data processing of the press release first publication device 200.

[0103] The input device 230 can be used to receive input numbers or signals. Among them, the signal can be a key signal generated in relation to the user settings and function control of the device / terminal / server. The output device 240 can include display devices such as display screens.

[0104] The above has described the embodiments of the present disclosure, the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes are obvious to those skilled in the art without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, practical application or technical improvement to the technology in the market, or to enable other ordinary skilled persons in the art to understand the embodiments disclosed herein.

Claims

1. A press release origination identification method, characterized by comprising: A method for identifying a first draft of an event in a news article, the method comprising: obtaining a news article to be analyzed, and extracting a title of the news article to be analyzed; searching for similar articles from a news article set according to search information composed of the title, search conditions, and a return number, specifically, a search engine obtains search information, and screens similar articles according to the obtained search information, when the number of news articles meeting the search information is less than the return number, the news articles meeting the search information are directly extracted as similar articles and returned, and when the number of news articles meeting the search information is greater than or equal to the return number, part of the news articles meeting the search conditions are extracted as similar articles and returned according to the return number; extracting entity words in the news article to be analyzed and entity words in the similar articles, respectively; matching the entity words in the news article to be analyzed with the entity words in the similar articles, and confirming whether the similar articles and the news article to be analyzed are articles of the same event according to a matching result; when the similar articles and the news article to be analyzed are articles of the same event, determining a first draft from the articles of the same event as the news article to be analyzed; when the similar articles are screened from the news article set, judging whether the number of returned similar articles is equal to the return number, if equal to the return number, updating and calibrating the screened similar articles first, and then performing subsequent operations such as extracting entity words in the news article to be analyzed and entity words in the similar articles, respectively, if less than the return number, directly performing subsequent operations such as extracting entity words in the news article to be analyzed and entity words in the similar articles, respectively; updating and calibrating the screened similar articles, including judging whether a minimum value of similarity between the similar articles and the title of the news article to be analyzed is less than a calibration threshold, when less than the calibration threshold, the screened similar articles do not need to be updated, and when greater than or equal to the calibration threshold, adjusting the return number in the search information, and then re-searching and updating and calibrating the similar articles based on the adjusted return number, until the number of returned similar articles is less than the adjusted return number or the minimum value of similarity between the returned similar articles and the title of the news article to be analyzed is less than the calibration threshold, and taking news articles in a final returned article list as similar articles of the news article to be analyzed after calibration.

2. The method of claim 1, wherein, when the similar articles are not screened from the news article set, directly identifying the news article to be analyzed as a first draft.

3. The method of claim 1, wherein, when the similar articles are screened from the news article set, further comprising: calculating and judging similarity between the news article to be analyzed and the similar articles; when it is judged that the similarity between the news article to be analyzed and the similar articles exceeds a preset threshold, performing the operation of extracting entity words in the news article to be analyzed and entity words in the similar articles, respectively.

4. The method of claim 3, wherein, When it is determined that the similarity between the news article to be analyzed and the similar article does not exceed the preset threshold, the news article to be analyzed is determined to be a first article.

5. The method of claim 3, wherein, When the similarity between the news article to be analyzed and the similar article is calculated, the calculation is based on a word vector model.

6. The method according to any one of claims 1-5, characterized in that, When the entity words in the news article to be analyzed and the entity words in the similar article are extracted, the extraction is based on a sequence labeling model.

7. The method according to any one of claims 1-5, characterized in that, When the entity words in the news article to be analyzed and the entity words in the similar article are matched, and whether the similar article and the news article to be analyzed are articles of the same event is determined according to the matching result, the following steps are included: When the same type of entity words are extracted from the entity words in the news article to be analyzed and the entity words in the similar article, consistency matching is performed. When there are the same type of entity words and there are consistent entity words in the same type of entity words, it is determined that the similar article and the news article to be analyzed are articles of the same event.

8. The method of claim 1, wherein, When each of the similar articles is not an article of the same event as the news article to be analyzed, the news article to be analyzed is determined to be a first article.

9. A news article origination identification apparatus characterized by comprising: The device for determining a first article of an event in a news article includes: A news article to be analyzed acquisition module is configured to acquire a news article to be analyzed and extract a title of the news article to be analyzed. A similar article screening module is configured to search an engine to screen similar articles from a news article set according to search information composed of the title, search conditions, and a return quantity. Specifically, the search engine acquires the search information, screens the similar articles according to the acquired search information, extracts all news articles that meet the search information as the similar articles when the number of news articles that meet the search information is less than the return quantity, and extracts part of the news articles that meet the search conditions as the similar articles when the number of news articles that meet the search information is greater than or equal to the return quantity. An entity word extraction module is configured to extract entity words in the news article to be analyzed and entity words in the similar articles when the similar articles are screened from the news article set. A same event article determination module is configured to match the entity words in the news article to be analyzed and the entity words in the similar articles, and determine whether the similar article and the news article to be analyzed are articles of the same event according to a matching result. A first article determination module is configured to determine a first article from the articles of the same event as the news article to be analyzed when the similar article and the news article to be analyzed are articles of the same event. When the similar articles are screened from the news article set, it is determined whether the number of returned similar articles is equal to the return quantity. If the number of returned similar articles is equal to the return quantity, the similar articles that are screened are updated and calibrated, and then subsequent operations such as extracting the entity words in the news article to be analyzed and the entity words in the similar articles are performed. If the number of returned similar articles is less than the return quantity, subsequent operations such as extracting the entity words in the news article to be analyzed and the entity words in the similar articles are directly performed. The updating and calibration of the screened similar articles comprises: judging whether the minimum value of the similarity between the title of the similar article and the title of the news article to be analyzed is less than the calibration threshold value; if the minimum value is less than the calibration threshold value, the screened similar article does not need to be supplemented and updated; if the minimum value is greater than or equal to the calibration threshold value, the return quantity in the search information is adjusted; then the search and updating and calibration of the similar article are re-performed based on the adjusted return quantity, until the number of the returned similar articles is less than the adjusted return quantity or the minimum value of the similarity between the returned similar article and the title of the news article to be analyzed is less than the calibration threshold value; and the news article in the finally returned article list is taken as the similar article of the calibrated news article to be analyzed.

10. A news article first publication identification apparatus characterized by comprising: The device for identifying the first article of an event in a news article comprises: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the method of any one of claims 1 to 8 when executing the executable instructions.

Images