CN-121615634-B - News topic selection generating method

CN121615634BCN 121615634 BCN121615634 BCN 121615634BCN-121615634-B

Abstract

The application relates to a news topic generation method which comprises the steps of obtaining multi-source data and generating event clusters, identifying candidate point sentences from the event clusters and generating structured data containing main bodies, contents, standpoints and source information, dividing the structured data into a plurality of layers based on identity characteristics and behavior patterns in the point source information and the main bodies, calculating credibility and event dynamic influence force of each layer, constructing scene vectors based on representative text contents of the event clusters, judging key layer deletion and calculating penalty factors, calculating value contributions of each layer based on the credibility and the event dynamic influence force, obtaining comprehensive layering value based on the value contributions of each layer and the penalty factors, constructing cross-layer relation signals based on the event clusters and the structured data, generating a plurality of candidate angles, and selecting the first N as final candidate angles according to comprehensive layering value score ordering.

Inventors

XIA TAO
ZHANG KAN
CHEN LIANG
Hang Keying

Assignees

浙江华智万像科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260203

Claims (8)

1. The news choice question generation method is characterized by comprising the following steps: Acquiring multi-source data and generating an event cluster based on the multi-source data, wherein the event cluster is an information aggregate of the same event; identifying candidate viewpoint sentences from a document set of the event cluster and generating structured data, wherein the structured data comprises viewpoint main bodies, viewpoint contents, viewpoint positions and viewpoint source information; dividing the structured data into a plurality of layers based on the identity characteristics and the behavior patterns in the viewpoint source information and the viewpoint main body, obtaining credibility based on long-term behavior data of the viewpoint main body for each layer, obtaining event dynamic influence based on index information associated with the multi-source data, wherein the layers comprise an authoritative layer, a professional layer, a social layer and a base layer; Inputting a representative text content of the event cluster into a multi-label classification model to generate a scene vector, determining a key level based on the scene vector, obtaining a key level missing judgment result by checking whether valid viewpoint data exists in the key level, and calculating a penalty factor based on the missing key level when the key level missing judgment result is missing, wherein the representative text content comprises a news text, an authoritative release abstract and a representative viewpoint content, the dimension of the scene vector comprises time administration, folk and public relations and industry science and technology or depth analysis, and the key level is one or more of the levels; calculating, for each tier, a tier value contribution based on the trustworthiness and the event dynamic impact; obtaining a comprehensive layering value based on the hierarchical value contributions of the various hierarchies and the penalty factors; Based on the cross-layer relation signal, the hierarchical value contribution and the structured data, generating a plurality of candidate angles; based on the comprehensive layering value, scoring and sorting the candidate angles, and selecting the top N candidate angles with the highest scores as final candidate angles; wherein constructing a cross-layer relationship signal based on the event clusters and the structured data comprises: Based on the viewpoint of the structured data corresponding to each hierarchy, obtaining a conflict signal; Based on the event clusters, calculating social layer discussion growth rate and authority layer content update strength to obtain a gap signal; obtaining an interoperability signal based on a representative viewpoint in the structured data corresponding to the professional layer and a representative viewpoint in the structured data corresponding to the base layer; The method comprises the steps of obtaining a multi-source data, calculating an original heat value based on index information associated with the multi-source data, obtaining a smooth heat value at a current moment based on the original heat value and a smooth heat value at a previous moment, and obtaining a trend signal based on the smooth heat value at the current moment, the smooth heat value at the previous moment and a trend sensitivity threshold; constructing a cross-layer relationship signal based on the conflict signal, the gap signal, the interoperability signal, and the trend signal; Wherein the generating a plurality of candidate angles based on the cross-layer relationship signal, the hierarchical value contribution, and the structured data comprises: based on the cross-layer relation signal and the hierarchical value contribution, respectively calculating signal trigger intensities corresponding to the conflict signal, the gap signal, the interoperability signal and the trend signal; And generating a plurality of candidate angles based on the plurality of topic angle signals, the structured data and the hierarchical value contribution, wherein the plurality of candidate angles respectively correspond to different types of cross-layer relation signals.
2. The news topic generation method of claim 1, wherein the obtaining the conflict signal based on the viewpoint in the structured data corresponding to each hierarchy includes: forming a standpoint distribution of the perspectives of each layer based on the standpoint in the structured data corresponding to each layer; Calculating the position difference degree between any two layers based on the position distribution; based on the standpoint difference, a conflict signal is obtained.
3. The news topic generation method of claim 1, wherein the obtaining an interoperability signal based on a representative view in the structured data corresponding to the professional layer and a representative view in the structured data corresponding to the base layer includes: Calculating semantic similarity based on the representative viewpoint in the structured data corresponding to the professional layer and the representative viewpoint in the structured data corresponding to the base layer; and obtaining an interoperability signal based on the semantic similarity and a preset credibility weight.
4. The news topic generation method of claim 1 wherein scoring the candidate angles based on the comprehensive hierarchical value and selecting the top N candidate angles as final candidate angles includes: Generating an innovation score based on the candidate angles; and comprehensively scoring the candidate angles based on the comprehensive layering value and the innovation degree score, and selecting the N candidate angles before comprehensive scoring as final candidate angles.
5. The news topic generation method of claim 4, wherein generating an innovation score based on the candidate angles includes: obtaining a content novelty score based on the candidate angle and the historical topic angle; calculating information entropy based on the candidate angles, and obtaining unique scores of the visual angles based on the information entropy; generating a formal innovation degree score based on the candidate angles and a preset narrative template library; and obtaining an innovative composite score based on the content novelty score, the visual angle unique score and the formal innovation score.
6. The news topic generation method of claim 1, wherein the obtaining the confidence level based on the long-term behavior data of the viewpoint main body in the multi-source data for each hierarchy includes: Calculating a historical score based on long-term behavior data of a viewpoint main body in the multi-source data for each level, and obtaining the credibility based on preset static credibility and the historical score; and/or, the obtaining the dynamic influence of the event based on the index information associated with the multi-source data includes: calculating the influence of the original event based on the index information associated with the multi-source data; extracting emotion polarization indexes, text homogenization and short-term abnormal burst characteristics based on the transmission content and transmission behavior data of the event clusters to obtain emotion type noise suppression factors; and obtaining the dynamic influence of the event based on the emotion type noise suppression factor and the original event influence.
7. The news topic generation method of claim 1 wherein the identifying candidate opinion sentences from the set of documents of the event cluster and generating structured data includes: Identifying viewpoint sentences from the document set of the event cluster to obtain candidate viewpoint sentences; inputting the candidate viewpoint sentences into a preset large language model to obtain initial structured data; Obtaining representative views and supporting evidence based on view content in the initial structured data; The structured data is generated based on the representative point of view, supporting evidence, point of view source information associated with the multi-source data, and the initial structured data.
8. The news topic generation method of any one of claims 1-7, further comprising, after the top N candidate angles with highest scores are selected as final candidate angles: Performing entity identification on the final candidate angle, and extracting a place name mechanism entity; Mapping the place name mechanism entity to a localized geographic knowledge graph, and calculating graph hop distance and semantic association degree of the place name mechanism entity and a local core node in the knowledge graph, wherein the knowledge graph comprises local administrative division nodes, pillar industry association nodes, key folk project nodes and place name alias mapping relations to form a structured knowledge network covering local core association elements; Based on the map hop count distance and the semantic association degree, judging the association level of the final candidate angle and the local core node; generating localized candidate angles based on the association level; and inputting the localization candidate angle into a preset executable scoring model, and evaluating space reachability, resource matching degree, time feasibility and news value score to generate an execution suggestion card.

Description

News topic selection generating method Technical Field The application relates to the technical field of news media, in particular to a news topic generation method. Background The television news topic serves as a core starting point of news production and directly determines the value guidance, the propagation effect and the audience acceptance of the report. At the moment of rapid development of new media technology and increasingly diverse information propagation, news clues are expanded from traditional media channels to full-network multi-source platforms such as government affair release, social media, industry report and the like, so that massive heterogeneous information ecology is formed. However, in the related art, shallow data such as multi-focus event popularity, topic flow, or public opinion emotion is statistically ranked, and it is difficult to deeply mine high-value topic angles. At present, an effective solution is not proposed for solving the problem that the high-value choice question angle is difficult to deeply excavate in the related technology. Disclosure of Invention The embodiment of the application provides a news topic generation method, which at least solves the problem that the high-value topic angle is difficult to deeply excavate in the related technology. The embodiment of the application provides a news choice question generation method, which comprises the following steps: Acquiring multi-source data and generating an event cluster based on the multi-source data, wherein the event cluster is an information aggregate of the same event; identifying candidate viewpoint sentences from a document set of the event cluster and generating structured data, wherein the structured data comprises viewpoint main bodies, viewpoint contents, viewpoint positions and viewpoint source information; Dividing the structured data into a plurality of layers based on the identity features and the behavior patterns in the viewpoint source information and the viewpoint main body, obtaining credibility based on long-term behavior data of the viewpoint main body for each layer, and obtaining event dynamic influence based on index information associated with the multi-source data; Identifying event propagation attributes based on the representative text content of the event clusters, and constructing scene vectors, obtaining a key-level missing judgment result based on the scene vectors, and calculating penalty factors based on the key-level missing judgment result; calculating, for each tier, a tier value contribution based on the trustworthiness and the event dynamic impact; obtaining a comprehensive layering value based on the hierarchical value contributions of the various hierarchies and the penalty factors; Based on the cross-layer relation signal, the hierarchical value contribution and the structured data, generating a plurality of candidate angles; and scoring and sorting the candidate angles based on the comprehensive layering value, and selecting the top N candidate angles with the highest scores as final candidate angles, wherein N is a positive integer. In some embodiments, the hierarchy comprises an authority layer, a professional layer, a social layer and a base layer, and the constructing a cross-layer relationship signal based on the event clusters and the structured data comprises: Based on the viewpoint of the structured data corresponding to each hierarchy, obtaining a conflict signal; Based on the event clusters, calculating social layer discussion growth rate and authority layer content update strength to obtain a gap signal; obtaining an interoperability signal based on a representative viewpoint in the structured data corresponding to the professional layer and a representative viewpoint in the structured data corresponding to the base layer; The method comprises the steps of obtaining a multi-source data, calculating an original heat value based on index information associated with the multi-source data, obtaining a smooth heat value at a current moment based on the original heat value and a smooth heat value at a previous moment, and obtaining a trend signal based on the smooth heat value at the current moment, the smooth heat value at the previous moment and a trend sensitivity threshold; And constructing a cross-layer relation signal based on the conflict signal, the gap signal, the interoperability signal and the trend signal. In some embodiments, the obtaining the conflict signal based on the viewpoint in the structured data corresponding to each hierarchy includes: forming a standpoint distribution of the perspectives of each layer based on the standpoint in the structured data corresponding to each layer; Calculating the position difference degree between any two layers based on the position distribution; based on the standpoint difference, a conflict signal is obtained. In some embodiments, the obtaining the interoperability signal based on the representative viewpoin