CN102646114A - News topic timeline abstract generating method based on breakthrough point - Google Patents
News topic timeline abstract generating method based on breakthrough point Download PDFInfo
- Publication number
- CN102646114A CN102646114A CN201210037970XA CN201210037970A CN102646114A CN 102646114 A CN102646114 A CN 102646114A CN 201210037970X A CN201210037970X A CN 201210037970XA CN 201210037970 A CN201210037970 A CN 201210037970A CN 102646114 A CN102646114 A CN 102646114A
- Authority
- CN
- China
- Prior art keywords
- topic
- news
- theme
- time slice
- breakthrough point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a news topic timeline abstract generating method based on a breakthrough point, and the method is used for automatically and efficiently discovering significant instants and important events happening in the development process of a target news topic, thereby greatly assisting readers in understanding the evolution process of a news topic. The method comprises the following steps: (1) according to topic keywords input by users, downloading all news articles (searched by using search terms) from related news websites, and then carrying out pretreatment on the news articles; (2) establishing a topic-activity hidden Markov model for the activity variation trend of a target topic in each time slice, and deleting the time slices in which the target topic is not active; (3) carrying out modeling on a topic shift sequence in each time slice by using a topic-shift hidden Markov model; (4) extracting sentences (relevant to an important event happening on that day) as an abstract of the breakthrough point; and (5) outputting the timeline abstract of the target topic.
Description
Technical field
The present invention relates to the technical field of Computer Applied Technology, relate to a kind of news topic timeline abstraction generating method particularly based on the breakthrough point.
Background technology
In the current information explosion epoch, people can freely read, download all kinds of news report about a news topic from the internet.Because the related news article quantity about a news topic (especially hot news topic) on the network is very many, cause reader be difficult to comform development trend and evolution process efficient, understanding target news topic in the news report of heterogeneous pass with saving time.
The difficult point of news topic timeline summary generation problem comprises how from the relevant news report of a news topic, confirming the material time point (being the breakthrough point) in this topic evolution, and how to make a summary according to the related news rise time line of a breakthrough point.Method of the prior art is described respectively below:
A) related work of breakthrough point excavation
Known method comprises four kinds at present, respectively as follows:
1. based on news quantity
This method is at first added up the related news quantity of target topic on each time point, and then that news topic quantity is maximum some time points are as the breakthrough point in this topic evolution.Use this method to obtain in all breakthrough points of a topic, it not is the material time point of this topic that a lot of breakthrough points are arranged.Because news report has the characteristics of " sudden-diversity ", the time point that therefore news report is many might not be exactly the material time point of this topic.
2. based on event monitoring
This method at first detects a series of media events relevant with the target topic the news article from each time point successively, and the time point that then each media event is taken place is as the breakthrough point of target topic.
Because most of dependent event of a news topic is not the major event in this topic development, the evolution process, therefore use this method to obtain in all breakthrough points of a topic, it not is the material time point of this topic that a lot of breakthrough points are arranged.
3. based on the fluctuation of emotion tendency
This method at first extracts people in emotion tendency and the intensity of this time point to the target topic the related article from each time point; Then through analyze people on each time point to the emotion tendency of target topic and the situation of strength fluctuation, excavate the breakthrough point of target topic.
Therefore this method is not suitable for and analyzes this main article type true, that subjectivity is very weak of describing of news report owing to rely on the emotion variation tendency of people to the target topic.
4. based on file correlation
This method is developed by Google company, and once is used to Google News Time line (Google's news timeline) network service.Because the realization details of this method is from unexposed, and Google News Timeb line service closed by Google company is permanent in July, 2011, therefore can think that this algorithm does not re-use.
B) related work of breakthrough point summary generation
Known method comprises three kinds at present, respectively as follows:
1. based on neural network
This method will be made a summary generative process with a self organizing neural network modeling, with the input of all summary candidate sentences as neural network, through the limit weight of continuous iterative computation neural network, finally export the sentence subclass of this breakthrough point summary.
2. based on graph structure
All sentences in the news article that this method was at first delivered the breakthrough point same day place one undirected weight graph to be arranged, and node is represented sentence, and two similarities between sentence are represented on the limit, and the weight on limit is the similarity size.From figure, select the summary of the sentence set of quantity of information maximum, redundance minimum through the random walk mode then as this breakthrough point.
3. based on optimized Algorithm
This method generative process of will making a summary is modeled as a linear optimization problem, and wherein each variable is represented a summary candidate sentence, and optimizing confined condition is the redundant restriction of sentence, and optimization aim is the maximum fault information that summary comprised.The sentence set that the breakthrough point summary was comprised when the mode through iteration optimization calculated the optimization aim convergence.
Above-mentioned three kinds of methods are not all considered the relevant major event of breakthrough point generation on the same day when generating the breakthrough point summary, but only consider to select the sentence that those contain much information and redundance is little, therefore can't guarantee that the summary and the breakthrough point itself that generate are closely related.
Listed related work more than comprehensive, as shown in Figure 1 based on the general flow of the news topic timeline abstraction generating method of breakthrough point.
Summary of the invention
For overcoming the defective of prior art, thereby the technical matters that the present invention will solve has provided the news topic timeline abstraction generating method based on the breakthrough point that a kind of major event of excavating the significant instant in the evolution of target news topic automatically, expeditiously and being taken place has greatly helped the evolution process of a news topic of reader in understanding.
Technical scheme of the present invention is: this news topic timeline abstraction generating method based on the breakthrough point may further comprise the steps:
(1) with the topic keyword of user input as term; Utilize reptile to download all news articles that obtain with the term search from the related news website; Then these news articles are carried out pre-service; Pre-service comprises: alphabetical small letterization, remove stop words, numeral and punctuation mark, and made up the news corpus of target topic thus;
(2) the liveness variation tendency of target topic on each time slice set up topic liveness HMM, and the sluggish time slice of deletion target topic;
(3) in each time slice that the target topic enlivens; At first utilize step (2) topic liveness HMM from the news language material of correspondence, to excavate each theme; Utilize the theme transition HMM that the theme transition sequence in each time slice is carried out modeling then; And calculate the intensity of each theme on each time point, through analyzing the strength fluctuation pattern of each theme on timeline, excavate the breakthrough point of target topic at last;
(4),, extract and the maximally related sentence of major event of generation on the same day summary as this breakthrough point through the degree of agreement of sentence in coupling subject key words and the news article to each breakthrough point of excavating;
(5) timeline of export target topic summary.
This method adopts the strategy of " dividing and rule "; Through the time division fragment, in each fragment, excavate breakthrough point and generate summary concurrently, make whole timeline summary generation scheme than common, to excavate the scheme of breakthrough point on the time cycle at whole topic more efficient, quick; This method is foundation with the topic fluctuation model but not serves as according to excavating the breakthrough point with newborn incident; Because the newborn incident that most of topic is relevant is not the critical event of this topic, thus this method excavate the breakthrough point accuracy rate will far above common, serve as according to the method for excavating the breakthrough point with newborn incident.
Description of drawings
Fig. 1 is the process flow diagram based on the news topic timeline abstraction generating method of breakthrough point of prior art;
Fig. 2 is the process flow diagram according to the news topic timeline abstraction generating method based on the breakthrough point of the present invention;
Fig. 3 is the process flow diagram of step of the present invention (2);
Fig. 4 is the process flow diagram of step of the present invention (3);
Fig. 5 is the process flow diagram of step of the present invention (4).
Embodiment
Do further detailed description in the face of technical scheme of the present invention down.
As shown in Figure 2, this news topic timeline abstraction generating method based on the breakthrough point may further comprise the steps:
(1) with the topic keyword of user input as term; Utilize reptile to download all news articles that obtain with the term search from the related news website; Then these news articles are carried out pre-service; Pre-service comprises: alphabetical small letterization, remove stop words, numeral and punctuation mark, and made up the news corpus of target topic thus;
(2) the liveness variation tendency of target topic on each time slice set up topic liveness HMM, and the sluggish time slice of deletion target topic;
(3) in each time slice that the target topic enlivens; At first utilize step (2) topic liveness HMM from the news language material of correspondence, to excavate each theme; Utilize the theme transition HMM that the theme transition sequence in each time slice is carried out modeling then; And calculate the intensity of each theme on each time point, through analyzing the strength fluctuation pattern of each theme on timeline, excavate the breakthrough point of target topic at last;
(4),, extract and the maximally related sentence of major event of generation on the same day summary as this breakthrough point through the degree of agreement of sentence in coupling subject key words and the news article to each breakthrough point of excavating;
(5) timeline of export target topic summary.
This method adopts the strategy of " dividing and rule "; Through the time division fragment, in each fragment, excavate breakthrough point and generate summary concurrently, make whole timeline summary generation scheme than common, to excavate the scheme of breakthrough point on the time cycle at whole topic more efficient, quick; This method is foundation with the topic fluctuation model but not serves as according to excavating the breakthrough point with newborn incident; Because the newborn incident that most of topic is relevant is not the critical event of this topic, thus this method excavate the breakthrough point accuracy rate will far above common, serve as according to the method for excavating the breakthrough point with newborn incident.
Preferably, as shown in Figure 3, step (2) comprises step by step following:
(2.1) time slice that is a plurality of designated length with the whole time cycle cutting of target topic;
(2.2) add up each time slice interior news article quantity and newly-added information amount respectively; The newly-added information amount is Ku Beike-Lai Bule Kullback-Leibler divergence value that this time slice word distributes and last time slice word distributes, and then the news number multiply by the correction news quantity of newly-added information amount as this time slice;
(2.3) utilize dynamic programming that the correction news quantity of each time slice is divided in several buckets; Require the maximum diffusibleness of all barrels minimum; The diffusibleness of a bucket refers in this barrel to revise maximal value and minimum value poor of news quantity; Calculate the mean value of revising news quantity in each barrel then, as the expectation news quantity of each time slice in this barrel;
(2.4) make up topic liveness HMM; Wherein considerable value is the news quantity in each time slice; Hide value and be each time slice corresponding topic liveness level; Emission probability is Poisson distribution, and transition probability is obtained by Bao Mu-Wei He Baum-Welch algorithm, utilizes Viterbi Viterbi algorithm computation then and exports the topic liveness level of each time slice.
Preferably, as shown in Figure 4, step (3) comprises step by step following:
(3.1) utilize topic liveness HMM from the corresponding news article of the active time slice of each topic, to excavate each theme;
(3.2) make up the theme transition HMM; Wherein considerable value is the word sequence that the word in each document is formed in this time slice; Hide the theme transition sequence that value constitutes for the corresponding theme of each word; Emission probability is the probability distribution that the topic in the topic model produces word, and transition probability is obtained by the Baum-Welch algorithm, utilizes the theme transition sequence in this time slice of Viterbi algorithm computation then;
(3.3) with word number that every day, theme produced and this day the likening to and be the intensity level of this theme of total words, calculate the intensity vector of the intensity level composition of all themes every day thus on the same day;
(3.4) with every day theme the Jansen-Shannon Jensen-Shannon divergence value of intensity vector of intensity vector and back one day theme as the theme strength fluctuation value of this day; If the theme strength fluctuation value of this day, is then judged the breakthrough point of this day for the target topic greater than proxima luce (prox. luc) and back one day theme strength fluctuation value; For in the time slice the earliest and those days the latest, if the intensity level of the maximum theme of this day intensity is then judged the breakthrough point of this day for the target topic greater than the mean value of the theme maximum intensity value of this time slice Nei Geri; Export the target topic breakthrough point of each time slice then.
Preferably, as shown in Figure 5, step (4) comprises step by step following:
(4.1) for each breakthrough point, extract the summary candidate sentence every piece of news article delivering from the same day, the summary candidate sentence comprises first of headline sentence and body;
(4.2) from each news candidate sentence, extract the information speech, the information speech comprises noun, verb, adjective and adverbial word;
(4.3) utilizing the information word set of each is a plurality of classifications with all candidate sentence clusters, and wherein distance metric adopts Jie Hade Jaccard similarity formula;
(4.4) from each classification, extract a representative sentences, the information speech quantity of representative sentences is maximum in this classification in each;
(4.5) investigate the Jaccard distance of set of words of information word set and this breakthrough point leitmotive of each representative sentences successively; And press this distance and from small to large all are represented the sentence ordering; Successively each representative sentences is joined in the summary of this breakthrough point by this rank sequencing then; Surpassed predetermined maximum length up to the length of this summary, wherein the length of summary is the total words that institute comprises representative sentences, and predetermined maximum length is maximum word numbers of making a summary and can comprise in a breakthrough point; The leitmotive word is the maximum set of letters of the maximum theme lower probability of this breakthrough point intensity level, arranges each breakthrough point summary and output then chronologically.
The above; It only is preferred embodiment of the present invention; Be not that the present invention is done any pro forma restriction, every foundation technical spirit of the present invention all still belongs to the protection domain of technical scheme of the present invention to any simple modification, equivalent variations and modification that above embodiment did.
Claims (4)
1. news topic timeline abstraction generating method based on the breakthrough point is characterized in that: may further comprise the steps:
(1) with the topic keyword of user input as term; Utilize reptile to download all news articles that obtain with the term search from the related news website; Then these news articles are carried out pre-service; Pre-service comprises: alphabetical small letterization, remove stop words, numeral and punctuation mark, and made up the news corpus of target topic thus;
(2) the liveness variation tendency of target topic on each time slice set up topic liveness HMM, and the sluggish time slice of deletion target topic;
(3) in each time slice that the target topic enlivens; At first utilize step (2) topic liveness HMM from the news language material of correspondence, to excavate each theme; Utilize the theme transition HMM that the theme transition sequence in each time slice is carried out modeling then; And calculate the intensity of each theme on each time point, through analyzing the strength fluctuation pattern of each theme on timeline, excavate the breakthrough point of target topic at last; The breakthrough point is concrete date, i.e. a time point;
(4),, extract and the maximally related sentence of major event of generation on the same day summary as this breakthrough point through the degree of agreement of sentence in coupling subject key words and the news article to each breakthrough point of excavating;
(5) timeline of export target topic summary, timeline refers in chronological sequence a plurality of breakthrough points of sequential organization.
2. the news topic timeline abstraction generating method based on the breakthrough point according to claim 1, it is characterized in that: step (2) comprises step by step following:
(2.1) time slice that is a plurality of designated length with the whole time cycle cutting of target topic;
(2.2) add up each time slice interior news article quantity and newly-added information amount respectively; The newly-added information amount is Ku Beike-Lai Bule Kullback-Leibler divergence value that this time slice word distributes and last time slice word distributes, and then the news number multiply by the correction news quantity of newly-added information amount as this time slice;
(2.3) utilize dynamic programming that the correction news quantity of each time slice is divided in several buckets; Require the maximum diffusibleness of all barrels minimum; The diffusibleness of a bucket refers in this barrel to revise maximal value and minimum value poor of news quantity; Calculate the mean value of revising news quantity in each barrel then, as the expectation news quantity of each time slice in this barrel;
(2.4) make up topic liveness HMM; Wherein considerable value is the news quantity in each time slice; Hide value and be each time slice corresponding topic liveness level; Emission probability is Poisson distribution, and transition probability is obtained by Bao Mu-Wei He Baum-Welch algorithm, utilizes Viterbi Viterbi algorithm computation then and exports the topic liveness level of each time slice.
3. the news topic timeline abstraction generating method based on the breakthrough point according to claim 2, it is characterized in that: step (3) comprises step by step following:
(3.1) utilize topic liveness HMM from the corresponding news article of the active time slice of each topic, to excavate each theme;
(3.2) make up the theme transition HMM; Wherein considerable value is the word sequence that the word in each document is formed in this time slice; Hide the theme transition sequence that value constitutes for the corresponding theme of each word; Emission probability is the probability distribution that the topic in the topic model produces word, and transition probability is obtained by the Baum-Welch algorithm, utilizes the theme transition sequence in this time slice of Viterbi algorithm computation then;
(3.3) with word number that every day, theme produced and this day the likening to and be the intensity level of this theme of total words, calculate the intensity vector of the intensity level composition of all themes every day thus on the same day;
(3.4) with every day theme the Jansen-Shannon Jensen-Shannon divergence value of intensity vector of intensity vector and back one day theme as the theme strength fluctuation value of this day; If the theme strength fluctuation value of this day, is then judged the breakthrough point of this day for the target topic greater than proxima luce (prox. luc) and back one day theme strength fluctuation value; For in the time slice the earliest and those days the latest, if the intensity level of the maximum theme of this day intensity is then judged the breakthrough point of this day for the target topic greater than the mean value of the theme maximum intensity value of this time slice Nei Geri; Export the target topic breakthrough point of each time slice then.
4. the news topic timeline abstraction generating method based on the breakthrough point according to claim 3, it is characterized in that: step (4) comprises step by step following:
(4.1) for each breakthrough point, extract the summary candidate sentence every piece of news article delivering from the same day, the summary candidate sentence comprises first of headline sentence and body;
(4.2) from each news candidate sentence, extract the information speech, the information speech comprises noun, verb, adjective and adverbial word;
(4.3) utilizing the information word set of each is a plurality of classifications with all candidate sentence clusters, and wherein distance metric adopts Jie Hade Jaccard similarity formula;
(4.4) from each classification, extract a representative sentences, the information speech quantity of representative sentences is maximum in this classification in each;
(4.5) investigate the Jaccard distance of set of words of information word set and this breakthrough point leitmotive of each representative sentences successively; And press this distance and from small to large all are represented the sentence ordering; Successively each representative sentences is joined in the summary of this breakthrough point by this rank sequencing then; Surpassed predetermined maximum length up to the length of this summary, wherein the length of summary is the total words that institute comprises representative sentences, and predetermined maximum length is maximum word numbers of making a summary and can comprise in a breakthrough point; The leitmotive word is the maximum set of letters of the maximum theme lower probability of this breakthrough point intensity level, arranges each breakthrough point summary and output then chronologically.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210037970XA CN102646114A (en) | 2012-02-17 | 2012-02-17 | News topic timeline abstract generating method based on breakthrough point |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210037970XA CN102646114A (en) | 2012-02-17 | 2012-02-17 | News topic timeline abstract generating method based on breakthrough point |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102646114A true CN102646114A (en) | 2012-08-22 |
Family
ID=46658933
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210037970XA Pending CN102646114A (en) | 2012-02-17 | 2012-02-17 | News topic timeline abstract generating method based on breakthrough point |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102646114A (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103116644A (en) * | 2013-02-26 | 2013-05-22 | 华南理工大学 | Method for mining orientation of Web themes and supporting decisions |
| CN103473263A (en) * | 2013-07-18 | 2013-12-25 | 大连理工大学 | News event development process-oriented visual display method |
| CN103500163A (en) * | 2013-07-24 | 2014-01-08 | 百度在线网络技术（北京）有限公司 | Method and device for recognizing event key progress |
| CN103942265A (en) * | 2014-03-26 | 2014-07-23 | 北京奇虎科技有限公司 | Method and device for pushing webpages containing news information |
| CN104182504A (en) * | 2014-08-18 | 2014-12-03 | 合肥工业大学 | Algorithm for dynamically tracking and summarizing news events |
| CN104484346A (en) * | 2014-11-28 | 2015-04-01 | 浙江大学 | Hierarchical theme modeling method based on mixed distance and relying on Chinese restaurant process |
| CN105787121A (en) * | 2016-03-25 | 2016-07-20 | 大连理工大学 | Microblog event abstract extracting method based on multiple storylines |
| CN106484724A (en) * | 2015-08-31 | 2017-03-08 | 富士通株式会社 | Information processor and information processing method |
| CN107273346A (en) * | 2016-03-30 | 2017-10-20 | 邻客音公司 | To the expansible excavation of popular opinion from text |
| CN107656997A (en) * | 2017-09-20 | 2018-02-02 | 广东欧珀移动通信有限公司 | Natural language processing method, apparatus, storage medium and terminal device |
| CN107688652A (en) * | 2017-08-31 | 2018-02-13 | 苏州大学 | The evolutionary abstraction generating method of Internet media event |
| CN108399241A (en) * | 2018-02-28 | 2018-08-14 | 福州大学 | A kind of emerging much-talked-about topic detecting system based on multiclass feature fusion |
| CN108701133A (en) * | 2016-11-30 | 2018-10-23 | 微软技术许可有限责任公司 | Recommendation is provided |
| CN108694183A (en) * | 2017-04-06 | 2018-10-23 | 北京国双科技有限公司 | A kind of search method and device |
| WO2018223718A1 (en) * | 2017-06-09 | 2018-12-13 | 平安科技（深圳）有限公司 | Trending topic detection method, apparatus and device, and medium |
| US10162870B2 (en) | 2015-09-30 | 2018-12-25 | International Business Machines Corporation | Historical summary visualizer for news events |
| CN109408782A (en) * | 2018-10-18 | 2019-03-01 | 中南大学 | Research hotspot based on KL distance similarity measurement develops behavioral value method |
| CN109522481A (en) * | 2018-11-07 | 2019-03-26 | 中山大学 | A kind of recommended method for expanding the user visual field based on Markov model |
| CN111475732A (en) * | 2020-04-13 | 2020-07-31 | 腾讯科技（深圳）有限公司 | Information processing method and device |
| CN111581967A (en) * | 2020-05-06 | 2020-08-25 | 西安交通大学 | News theme event detection method combining LW2V and triple network |
| CN112612944A (en) * | 2020-12-07 | 2021-04-06 | 深圳价值在线信息科技股份有限公司 | Case information management method, terminal equipment and system |
| CN113254632A (en) * | 2021-04-22 | 2021-08-13 | 国家计算机网络与信息安全管理中心 | Timeline abstract automatic generation method based on event detection technology |
| CN113918708A (en) * | 2021-12-15 | 2022-01-11 | 深圳市迪博企业风险管理技术有限公司 | Abstract extraction method |
-
2012
- 2012-02-17 CN CN201210037970XA patent/CN102646114A/en active Pending
Non-Patent Citations (1)
| Title |
|---|
| PO HU ETC.: "Generating Breakpoint-based Timeline Overview for News Topic Retrospection", 《ICDM 2011 11TH IEEE INTERNATIONAL CONFERENCE ON DATA MINING》 * |
Cited By (37)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103116644A (en) * | 2013-02-26 | 2013-05-22 | 华南理工大学 | Method for mining orientation of Web themes and supporting decisions |
| CN103116644B (en) * | 2013-02-26 | 2016-04-13 | 华南理工大学 | Web topic tendentiousness excavates the method with decision support |
| CN103473263A (en) * | 2013-07-18 | 2013-12-25 | 大连理工大学 | News event development process-oriented visual display method |
| CN103473263B (en) * | 2013-07-18 | 2017-02-08 | 大连理工大学 | News event development process-oriented visual display method |
| CN103500163B (en) * | 2013-07-24 | 2016-12-28 | 百度在线网络技术（北京）有限公司 | The method and apparatus of identification event key development |
| CN103500163A (en) * | 2013-07-24 | 2014-01-08 | 百度在线网络技术（北京）有限公司 | Method and device for recognizing event key progress |
| CN103942265A (en) * | 2014-03-26 | 2014-07-23 | 北京奇虎科技有限公司 | Method and device for pushing webpages containing news information |
| CN104182504A (en) * | 2014-08-18 | 2014-12-03 | 合肥工业大学 | Algorithm for dynamically tracking and summarizing news events |
| CN104182504B (en) * | 2014-08-18 | 2017-06-06 | 合肥工业大学 | A kind of dynamic tracking of media event and summary algorithm |
| CN104484346A (en) * | 2014-11-28 | 2015-04-01 | 浙江大学 | Hierarchical theme modeling method based on mixed distance and relying on Chinese restaurant process |
| CN104484346B (en) * | 2014-11-28 | 2018-02-09 | 浙江大学 | A kind of stratification theme modeling method that Chinese-style restaurant's process is relied on based on mixing distance |
| CN106484724A (en) * | 2015-08-31 | 2017-03-08 | 富士通株式会社 | Information processor and information processing method |
| US10162870B2 (en) | 2015-09-30 | 2018-12-25 | International Business Machines Corporation | Historical summary visualizer for news events |
| CN105787121A (en) * | 2016-03-25 | 2016-07-20 | 大连理工大学 | Microblog event abstract extracting method based on multiple storylines |
| CN105787121B (en) * | 2016-03-25 | 2018-08-14 | 大连理工大学 | A kind of microblogging event summary extracting method based on more story lines |
| CN107273346A (en) * | 2016-03-30 | 2017-10-20 | 邻客音公司 | To the expansible excavation of popular opinion from text |
| CN107273346B (en) * | 2016-03-30 | 2024-06-11 | 微软技术许可有限责任公司 | Extensible mining of trending insights from text |
| CN108701133A (en) * | 2016-11-30 | 2018-10-23 | 微软技术许可有限责任公司 | Recommendation is provided |
| US11494450B2 (en) | 2016-11-30 | 2022-11-08 | Microsoft Technology Licensing, Llc | Providing recommended contents |
| CN108694183A (en) * | 2017-04-06 | 2018-10-23 | 北京国双科技有限公司 | A kind of search method and device |
| WO2018223718A1 (en) * | 2017-06-09 | 2018-12-13 | 平安科技（深圳）有限公司 | Trending topic detection method, apparatus and device, and medium |
| CN107688652B (en) * | 2017-08-31 | 2020-12-29 | 苏州大学 | Evolution type abstract generation method facing internet news events |
| CN107688652A (en) * | 2017-08-31 | 2018-02-13 | 苏州大学 | The evolutionary abstraction generating method of Internet media event |
| CN107656997A (en) * | 2017-09-20 | 2018-02-02 | 广东欧珀移动通信有限公司 | Natural language processing method, apparatus, storage medium and terminal device |
| CN107656997B (en) * | 2017-09-20 | 2021-01-15 | Oppo广东移动通信有限公司 | Natural language processing method and device, storage medium and terminal equipment |
| CN108399241B (en) * | 2018-02-28 | 2021-08-31 | 福州大学 | Emerging hot topic detection system based on multi-class feature fusion |
| CN108399241A (en) * | 2018-02-28 | 2018-08-14 | 福州大学 | A kind of emerging much-talked-about topic detecting system based on multiclass feature fusion |
| CN109408782A (en) * | 2018-10-18 | 2019-03-01 | 中南大学 | Research hotspot based on KL distance similarity measurement develops behavioral value method |
| CN109522481A (en) * | 2018-11-07 | 2019-03-26 | 中山大学 | A kind of recommended method for expanding the user visual field based on Markov model |
| CN111475732A (en) * | 2020-04-13 | 2020-07-31 | 腾讯科技（深圳）有限公司 | Information processing method and device |
| CN111581967A (en) * | 2020-05-06 | 2020-08-25 | 西安交通大学 | News theme event detection method combining LW2V and triple network |
| CN111581967B (en) * | 2020-05-06 | 2023-08-11 | 西安交通大学 | News theme event detection method combining LW2V with triple network |
| CN112612944A (en) * | 2020-12-07 | 2021-04-06 | 深圳价值在线信息科技股份有限公司 | Case information management method, terminal equipment and system |
| CN112612944B (en) * | 2020-12-07 | 2024-05-31 | 深圳价值在线信息科技股份有限公司 | Case information management method, terminal equipment and system |
| CN113254632A (en) * | 2021-04-22 | 2021-08-13 | 国家计算机网络与信息安全管理中心 | Timeline abstract automatic generation method based on event detection technology |
| CN113254632B (en) * | 2021-04-22 | 2022-07-22 | 国家计算机网络与信息安全管理中心 | Timeline abstract automatic generation method based on event detection technology |
| CN113918708A (en) * | 2021-12-15 | 2022-01-11 | 深圳市迪博企业风险管理技术有限公司 | Abstract extraction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102646114A (en) | News topic timeline abstract generating method based on breakthrough point | |
| US11487945B2 (en) | Predictive similarity scoring subsystem in a natural language understanding (NLU) framework | |
| US10713441B2 (en) | Hybrid learning system for natural language intent extraction from a dialog utterance | |
| Cao et al. | Knowledge-enriched event causality identification via latent structure induction networks | |
| US11520992B2 (en) | Hybrid learning system for natural language understanding | |
| CN103150382B (en) | Automatic short text semantic concept expansion method and system based on open knowledge base | |
| US20140214399A1 (en) | Translating natural language descriptions to programs in a domain-specific language for spreadsheets | |
| CN108563734A (en) | Institutional information querying method, device, computer equipment and storage medium | |
| Lamela Seijas et al. | Towards property-based testing of restful web services | |
| Leonandya et al. | A semi-supervised algorithm for Indonesian named entity recognition | |
| Vo | Se4exsum: An integrated semantic-aware neural approach with graph convolutional network for extractive text summarization | |
| CN103927176A (en) | Method for generating program feature tree on basis of hierarchical topic model | |
| CN110555199B (en) | Article generation method, device, equipment and storage medium based on hotspot materials | |
| CN102982063A (en) | Control method based on tuple elaboration of relation keywords extension | |
| CN104572111B (en) | A kind of program comprehension and characteristic positioning method based on related subject model | |
| Basile et al. | Entity linking for tweets | |
| CN113536772A (en) | Text processing method, device, equipment and storage medium | |
| Uddin et al. | A neural network approach for Bangla POS tagger | |
| Domingo Roig et al. | Enhancing sequence-to-sequence modeling for RDF triples to natural text | |
| CN115455155B (en) | Method for extracting subject information of government affair text and storage medium | |
| Fang et al. | Adaptive Code Completion with Meta-learning | |
| Long et al. | Deep Neural Network with Embedding Fusion for Chinese Named Entity Recognition | |
| CN115408353A (en) | Log data processing method and device | |
| Wang et al. | CACV-tree: A New Computational Approach for Sentence Similarity Modeling | |
| Cohn et al. | Ranking Content based on Semantic Dimensions: A Multi-objective Approach |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20120822 |