Tracking Method and Device for News Events, Storage Medium, Electronic Device

Abstract

The present disclosure relates to a method and apparatus for tracking news events, a storage medium, and an electronic device, and relates to the technical field of big data processing. The method includes: matching a second interest tag corresponding to a first interest tag in the topic level of a preset two-layer general concept topic tree; in the keyword level under the second interest tag of the preset two-layer general concept topic tree, matching a second keyword corresponding to a first keyword of the current news event; determining a mounting position of the current news event in the preset two-layer general concept topic tree according to a first-level position of the second interest tag in the topic level and a second-level position of the second keyword in the keyword level; and mounting the current news event to the preset two-layer general concept topic tree according to the mounting position, so that the current news event can be tracked according to the second interest tag and / or the second keyword. The present disclosure improves the mounting efficiency of the current news event.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the technical field of big data processing, and more specifically, embodiments of the present disclosure relate to a method for tracking news events, an apparatus for tracking news events, a computer-readable storage medium, and an electronic device. Background Art

[0002] This section aims to provide background or context for the embodiments of the present disclosure recited in the claims. The description herein is not admitted to be prior art merely by virtue of being included in this section.

[0003] In some news event tracking solutions, the following technical solutions can be implemented: by calculating the similarity between news events, performing calculations on all news events through offline clustering, and dividing each news event into a suitable cluster to achieve the purpose of tracking news events.

[0004] However, the above solutions cannot cluster news events reported in real time, and thus cannot track news events reported in real time. Summary of the Invention

[0005] Therefore, there is a great need for an improved method for tracking news events to mount the current news event to a two-layer general concept topic tree generated based on historical news events according to the interest tags and keywords of the current news event, so that the current news event can be tracked according to the interest tags and / or keywords.

[0006] In this context, embodiments of the present disclosure are expected to provide a method for tracking news events, an apparatus for tracking news events, a computer-readable storage medium, and an electronic device.

[0007] According to one aspect of the present disclosure, there is provided a method for tracking news events, including:

[0008] Extracting a first interest tag of the current news event and matching a second interest tag corresponding to the first interest tag in the topic hierarchy of a preset two-layer general concept topic tree; wherein, the preset two-layer general concept topic tree is generated based on historical news events;

[0009] Matching a second keyword corresponding to the first keyword of the current news event in the keyword hierarchy under the second interest tag of the preset two-layer general concept topic tree;

[0010] Determining a mounting position of the current news event in the preset two-layer general concept topic tree according to a first layer position of the second interest tag in the topic hierarchy and a second layer position of the second keyword in the keyword hierarchy;

[0011] According to the mounting position, mount the current news event to the preset two - layer general concept topic tree to track the current news event according to the second interest tag and / or the second keyword.

[0012] In an exemplary embodiment of the present disclosure, determining the mounting position of the current news event in the preset two - layer general concept topic tree according to the first - level position of the second interest tag in the topic hierarchy and the second - level position of the second keyword in the keyword hierarchy includes:

[0013] When it is determined that there is a second keyword corresponding to the first keyword in the keyword hierarchy, determine the mounting position of the current news event in the preset two - layer general concept topic tree according to the first - level position of the second interest tag in the topic hierarchy and the second - level position of the second keyword in the keyword hierarchy.

[0014] In an exemplary embodiment of the present disclosure, the news event tracking method further includes:

[0015] When it is determined that there is no second keyword corresponding to the first keyword in the keyword hierarchy, add the first keyword to the keyword hierarchy;

[0016] Determine the mounting position of the current news event in the preset two - layer general concept topic tree according to the first - level position of the second interest tag in the topic hierarchy and the second - level position of the first keyword in the keyword hierarchy.

[0017] In an exemplary embodiment of the present disclosure, after determining the mounting position of the current news event in the preset two - layer general concept topic tree, the news event tracking method further includes:

[0018] Calculate the similarity between the current news event and the historical news events corresponding to the second keyword;

[0019] When it is determined that the similarity is greater than a first preset threshold, merge the current news event and the historical news events corresponding to the second keyword, and update the historical news events corresponding to the second keyword with the merged news event.

[0020] In an exemplary embodiment of the present disclosure, calculating the similarity between the current news event and the historical news events corresponding to the second keyword includes:

[0021] Input the current news event into the first Sentence-Bert model to obtain a first vector representation, and input the historical news event corresponding to the second keyword into the second Sentence-Bert model to obtain a second vector representation;

[0022] Calculate the cosine similarity between the first vector representation and the second vector representation to obtain the similarity between the current news event and the historical news event corresponding to the second keyword.

[0023] In an exemplary embodiment of the present disclosure, the news event tracking method further includes:

[0024] Obtain historical news events, and use a preset interest tag extraction model to extract second interest tags of the historical news events;

[0025] According to the second interest tags and the first co-occurrence relationship between the second interest tags, construct a first network graph, and segment the first network graph to obtain a plurality of first sub-graph segmentation results; wherein, in each of the first sub-graph segmentation results, a plurality of second interest tags are included;

[0026] Construct a keyword set according to the second keywords of the historical news events included in each of the first sub-graph segmentation results, and construct a second network graph according to the second keywords included in the keyword set and the second co-occurrence relationship between the second keywords;

[0027] Segment the second network graph to obtain a plurality of second sub-graph segmentation results, and generate the preset two-layer general concept topic tree according to the first sub-graph segmentation results and the second sub-graph segmentation results.

[0028] In an exemplary embodiment of the present disclosure, constructing a first network graph according to the second interest tags and the first co-occurrence relationship between the second interest tags includes:

[0029] Perform abstraction processing on the second interest tags and the first co-occurrence relationship between the second interest tags to obtain vertices and edges;

[0030] Calculate the first co-occurrence times of the second interest tags in the same historical news event, and perform abstraction processing on the first co-occurrence times to obtain the weights of the edges;

[0031] Construct a first network graph based on the vertices, edges, and weights of the edges.

[0032] In an exemplary embodiment of the present disclosure, segmenting the first network graph to obtain a plurality of first sub-graph segmentation results includes:

[0033] S1. Establish a split / coalescence tree corresponding to the vertices included in the first network graph based on a preset community discovery algorithm;

[0034] S2. Calculate the target weight of each subgraph segmentation result according to the degree of each vertex and the number of expected first subgraph segmentation results; wherein, the degree of the vertex is the sum of the weights of the edges connected to the vertex;

[0035] S3. Traverse the split / coalescence tree in a preset order, and determine whether the degree of the vertices under the tree nodes included in the split / coalescence tree is greater than the target weight;

[0036] S4. If the degree of the vertices under any tree node is greater than the target weight, further search the branches of this tree node; otherwise, stop the search and use the vertex set corresponding to this tree node as a first subgraph segmentation result;

[0037] S5. Repeat steps S4 and S5 until the total weight of the vertices under all tree nodes is less than or equal to the target weight.

[0038] In an exemplary embodiment of the present disclosure, the news event tracking method further includes:

[0039] Calculate the first modularity of each of the first subgraph segmentation results, and evaluate the first subgraph segmentation results according to the first modularity.

[0040] In an exemplary embodiment of the present disclosure, the news event tracking method further includes:

[0041] Calculate the number of historical news events and / or current news events included under keywords of different categories at the keyword level;

[0042] When it is determined that the number of historical news events and / or current news events included under the keywords of any category is greater than a second preset threshold, split the historical news events and / or current news events included under the keywords of this category, and update the keywords of this category according to the split results.

[0043] According to one aspect of the present disclosure, there is provided a news event tracking device, including:

[0044] A first matching module, configured to extract a first interest tag of a current news event, and match a second interest tag corresponding to the first interest tag in the topic hierarchy of a preset two-layer general concept topic tree; wherein, the preset two-layer general concept topic tree is generated based on historical news events;

[0045] A second matching module, configured to match a second keyword corresponding to the first keyword of the current news event in the keyword hierarchy under the second interest tag;

[0046] A mounting position determination module, configured to determine a mounting position of the current news event in the preset two - layer general concept topic tree according to a first - layer position of the second interest tag in the topic hierarchy and a second - layer position of the second keyword in the keyword hierarchy;

[0047] An event tracking module, configured to mount the current news event to the preset two - layer general concept topic tree according to the mounting position, so as to track the current news event according to the second interest tag and / or the second keyword.

[0048] In an exemplary embodiment of the present disclosure, determining a mounting position of the current news event in the preset two - layer general concept topic tree according to a first - layer position of the second interest tag in the topic hierarchy and a second - layer position of the second keyword in the keyword hierarchy includes:

[0049] When it is determined that there is a second keyword corresponding to the first keyword in the keyword hierarchy, determining a mounting position of the current news event in the preset two - layer general concept topic tree according to a first - layer position of the second interest tag in the topic hierarchy and a second - layer position of the second keyword in the keyword hierarchy.

[0050] In an exemplary embodiment of the present disclosure, the news event tracking device further includes:

[0051] A first adding module, configured to add the first keyword to the keyword hierarchy when it is determined that there is no second keyword corresponding to the first keyword in the keyword hierarchy;

[0052] A first position determination module, configured to determine a mounting position of the current news event in the preset two - layer general concept topic tree according to a first - layer position of the second interest tag in the topic hierarchy and a second - layer position of the first keyword in the keyword hierarchy.

[0053] In an exemplary embodiment of the present disclosure, the news event tracking device further includes:

[0054] A first similarity calculation module, configured to calculate a similarity between the current news event and a historical news event corresponding to the second keyword;

[0055] An event merging module, configured to merge the current news event and the historical news event corresponding to the second keyword when it is determined that the similarity is greater than a first preset threshold, and update the historical news event corresponding to the second keyword by using the merged news event.

[0056] In an exemplary embodiment of the present disclosure, calculating the similarity between the current news event and the historical news event corresponding to the second keyword includes:

[0057] Inputting the current news event into a first Sentence - Bert model to obtain a first vector representation, and inputting the historical news event corresponding to the second keyword into a second Sentence - Bert model to obtain a second vector representation;

[0058] Calculating the cosine similarity between the first vector representation and the second vector representation to obtain the similarity between the current news event and the historical news event corresponding to the second keyword.

[0059] In an exemplary embodiment of the present disclosure, the news event tracking device further includes:

[0060] An interest tag extraction module, configured to obtain historical news events and use a preset interest tag extraction model to extract second interest tags of the historical news events;

[0061] A first graph segmentation module, configured to construct a first network graph according to the second interest tags and a first co - occurrence relationship between the second interest tags, and segment the first network graph to obtain a plurality of first sub - graph segmentation results; wherein, each of the first sub - graph segmentation results includes a plurality of second interest tags;

[0062] A network graph construction module, configured to construct a keyword set according to the second keywords of the historical news events included in each of the first sub - graph segmentation results, and construct a second network graph according to the second keywords included in the keyword set and a second co - occurrence relationship between the second keywords;

[0063] A second graph segmentation module, configured to segment the second network graph to obtain a plurality of second sub - graph segmentation results, and generate the preset two - layer general concept topic tree according to the first sub - graph segmentation results and the second sub - graph segmentation results.

[0064] In an exemplary embodiment of the present disclosure, constructing a first network graph according to the second interest tags and a first co - occurrence relationship between the second interest tags includes:

[0065] Abstract the second interest tags and the first co-occurrence relationships between the second interest tags to obtain vertices and edges;

[0066] Calculate the first co-occurrence times of the second interest tags in the same historical news event, and abstract the first co-occurrence times to obtain the weights of the edges;

[0067] Construct a first network graph based on the vertices, edges, and the weights of the edges.

[0068] In an exemplary embodiment of the present disclosure, segment the first network graph to obtain multiple first sub-graph segmentation results, including:

[0069] S1, establish a split / coalescence tree corresponding to the vertices included in the first network graph based on a preset community discovery algorithm;

[0070] S2, calculate the target weights of each sub-graph segmentation result according to the degrees of the vertices and the number of the first sub-graph segmentation results expected to be obtained; wherein, the degree of the vertex is the sum of the weights of the edges connected to the vertex;

[0071] S3, traverse the split / coalescence tree in a preset order, and judge whether the degree of the vertices under the tree nodes included in the split / coalescence tree is greater than the target weight;

[0072] S4, if the degree of the vertices under any tree node is greater than the target weight, further search the branches of the tree node; otherwise, stop the search, and use the vertex set corresponding to the tree node as a first sub-graph segmentation result;

[0073] S5, repeat steps S4 and S5 until the total weights of the vertices under all tree nodes are less than or equal to the target weight.

[0074] In an exemplary embodiment of the present disclosure, the tracking device for news events further includes:

[0075] An event quantity calculation module, configured to calculate the quantities of historical news events and / or current news events included under keywords of different categories at the keyword level;

[0076] An event splitting module, configured to split the historical news events and / or current news events included under the keywords of any category when it is determined that the quantity of the historical news events and / or current news events included under the keywords of any category is greater than a second preset threshold, and update the keywords of the category according to the splitting result.

[0077] According to one aspect of the present disclosure, there is provided a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the tracking method of the news event described in any one of the above is implemented.

[0078] According to one aspect of the present disclosure, there is provided an electronic device, including:

[0079] a processor; and

[0080] a memory for storing executable instructions of the processor;

[0081] wherein the processor is configured to execute the tracking method of the news event described in any one of the above by executing the executable instructions.

[0082] The tracking method of the news event and the tracking device of the news event according to the embodiments of the present disclosure can extract the first interest tag of the current news event, and match the second interest tag corresponding to the first interest tag in the topic level of the preset two-layer general concept topic tree; and in the keyword level under the second interest tag of the preset two-layer general concept topic tree, match the second keyword corresponding to the first keyword of the current news event; then determine the mounting position of the current news event in the preset two-layer general concept topic tree according to the first-level position of the second interest tag in the topic level and the second-level position of the second keyword in the keyword level; finally, mount the current news event to the preset two-layer general concept topic tree according to the mounting position, so that the current news event can be tracked according to the second interest tag and / or the second keyword, without clustering all news events, thereby significantly reducing the problem of low clustering efficiency caused by clustering all news events and the problem of being unable to cluster news events reported in real time, and reducing the problem of being unable to track news events reported in real time according to the keywords of the already-clustered news events due to the inability to associate the already-clustered news events with the news events reported in real time, bringing a better experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present disclosure will become readily understood. In the drawings, several embodiments of the present disclosure are shown by way of example and not by way of limitation, wherein:

[0084] Figure 1 Schematically shows a flowchart of a method for tracking a news event according to an exemplary embodiment of the present disclosure;

[0085] Figure 2 Schematically shows a structural example diagram of a general concept topic tree according to an exemplary embodiment of the present disclosure;

[0086] Figure 3 Schematically shows a flowchart of a method for generating a two - layer general - concept topic tree according to an exemplary embodiment of the present disclosure;

[0087] Figure 4 Schematically shows a flowchart of a method for dividing the first network graph to obtain multiple first sub - graph division results according to an exemplary embodiment of the present disclosure;

[0088] Figure 5 Schematically shows a flowchart of another method for tracking news events according to an exemplary embodiment of the present disclosure;

[0089] Figure 6 Schematically shows a structural example diagram of a similarity calculation model according to an exemplary embodiment of the present disclosure;

[0090] Figure 7 Schematically shows a flowchart of another method for tracking news events according to an exemplary embodiment of the present disclosure;

[0091] Figure 8 Schematically shows a block diagram of a news event tracking device according to an exemplary embodiment of the present disclosure;

[0092] Figure 9 Schematically shows a computer - readable storage medium for storing the above - mentioned news event tracking method according to an exemplary embodiment of the present disclosure;

[0093] Figure 10 Schematically shows an electronic device for implementing the above - mentioned news event tracking method according to an exemplary embodiment of the present disclosure.

[0094] In the drawings, the same or corresponding reference numerals indicate the same or corresponding parts. Detailed implementation manners

[0095] Hereinafter, the principles and spirit of the present disclosure will be described with reference to several exemplary embodiments. It should be understood that these embodiments are given only to enable those skilled in the art to better understand and then implement the present disclosure, and do not limit the scope of the present disclosure in any way. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to be able to fully convey the scope of the present disclosure to those skilled in the art.

[0096] Those skilled in the art know that the embodiments of the present disclosure can be implemented as a system, a device, an apparatus, a method, or a computer program product. Therefore, the present disclosure can be specifically implemented in the following forms: completely hardware, completely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0097] According to an embodiment of the present disclosure, a method for tracking news events, a device for tracking news events, a computer-readable storage medium, and an electronic device are provided.

[0098] In this article, the number of any element in the drawings is for illustration rather than limitation, and any naming is only for distinction without any limiting meaning.

[0099] Next, with reference to several representative embodiments of the present disclosure, the principles and spirit of the present disclosure will be explained in detail. Summary of the Invention

[0101] The inventor first considered that the topic to which a news event belongs can be identified by machine learning methods, and generally a supervised classification method is adopted; specifically, the supervised classification method pre-sets event categories, such as major social events, major policies and regulations, etc., and then trains a classifier corresponding to each type based on the event category; at the same time, in order to improve the accuracy of the classification result, some solutions will also pre-set seed words and dynamically increase the seed words by means of semantic expansion to achieve the effect of monitoring hot events.

[0102] However, since the event types are pre-set to train the classifier, the types that can be monitored are limited, and there is a problem of low news coverage; moreover, pre-setting types requires a large amount of human cost to select suitable topics, and a labeling team is required to label the news that conforms to the topic, and the development waiting time is relatively long. It is only suitable for the mining and tracking of long-term topics. The real-time hotspots change rapidly, and it is difficult to monitor this method.

[0103] Based on this, some researchers have proposed an unsupervised clustering solution, which can calculate the similarity between news events and use a clustering algorithm to divide the reports into multiple clusters, and each clustering cluster can be regarded as a topic; then, offline clustering calculates the full amount of data and divides the data into appropriate clusters.

[0104] However, traditional offline clustering methods use batch-learning, such as kmeans, dbscan, ap, hierarchical clustering, etc., and cannot cluster real-time news events in time. Due to the lack of data guidance, the accuracy is lower than that of supervised classification schemes, and the clustering results are not stable enough; moreover, limited by the model complexity and storage space, the large-scale offline data clustering takes a long time, cannot respond to real-time data, and cannot associate the clustered articles with real-time articles.

[0105] Meanwhile, if online clustering based on streaming computing is used to add new news to existing topic clusters for topic detection and tracking, although the above problems can be solved, the accuracy of the addition results is relatively low, making it impossible to accurately track real-time news events based on existing topic clusters.

[0106] Based on the above problems, the exemplary embodiments of the present disclosure first provide a method for tracking news events. On the one hand, since the mounting position of the current news event in the two-layer general concept topic tree can be determined according to the first interest tag and the first keyword of the current news event, and then the current news event can be mounted to the preset two-layer general concept topic tree according to the mounting position, so that the current news event can be tracked according to the second interest tag and / or the second keyword, without the need to classify the current news event through a classifier, thereby avoiding the problem of setting event types to train the classifier, resulting in a limited monitorable type and a low news coverage, and the problem of wasting time cost caused by the need to manually select appropriate topics and label news that conforms to the topics; on the other hand, it can also avoid the problem that the accuracy of the classification results is relatively low due to the lack of data guidance; on the further hand, the second interest tag and the second keyword corresponding to the first interest tag and the first keyword can be directly matched, and then the current news event can be associated with historical news events, solving the problem in the prior art that clustered articles and real-time articles cannot be associated.

[0107] After introducing the basic principle of the present disclosure, the various non-limiting implementation manners of the present disclosure will be specifically introduced below.

[0108] Exemplary Method

[0109] In this exemplary embodiment, a method for tracking news events is first provided. This method can run on a server, a server cluster, a cloud server, etc.; of course, those skilled in the art can also run the method of the present disclosure on other platforms according to requirements, and no special limitation is made in this exemplary embodiment. Refer to Figure 1 As shown, the method for tracking news events may include the following steps:

[0110] Step S110. Extract the first interest tag of the current news event, and match the second interest tag corresponding to the first interest tag in the topic hierarchy of the preset two-layer general concept topic tree; wherein, the preset two-layer general concept topic tree is generated according to historical news events;

[0111] Step S120. In the keyword hierarchy under the second interest tag of the preset two-layer general concept topic tree, match the second keyword corresponding to the first keyword of the current news event;

[0112] Step S130. Determine the mounting position of the current news event in the preset two - layer general - concept topic tree according to the first - layer position of the second interest tag in the topic hierarchy and the second - layer position of the second keyword in the keyword hierarchy.

[0113] Step S140. Mount the current news event to the preset two - layer general - concept topic tree according to the mounting position, so as to track the current news event based on the second interest tag and / or the second keyword.

[0114] In the above - mentioned news event tracking method, the first interest tag of the current news event can be extracted, and the second interest tag corresponding to the first interest tag is matched in the topic hierarchy of the preset two - layer general - concept topic tree; and in the keyword hierarchy under the second interest tag of the preset two - layer general - concept topic tree, the second keyword corresponding to the first keyword of the current news event is matched; then, according to the first - layer position of the second interest tag in the topic hierarchy and the second - layer position of the second keyword in the keyword hierarchy, the mounting position of the current news event in the preset two - layer general - concept topic tree is determined; finally, according to the mounting position, the current news event is mounted to the preset two - layer general - concept topic tree, so that the current news event can be tracked based on the second interest tag and / or the second keyword, without performing offline clustering on all news events. Thus, the problem of low clustering efficiency caused by offline clustering of all news events and the inability to perform real - time clustering on real - time reported news events is significantly reduced, and the problem of being unable to perform real - time tracking on real - time reported news events based on the keywords of the already - clustered news events due to the inability to associate the already - clustered news events with real - time reported news events is reduced, bringing a better experience to users.

[0115] Hereinafter, the news event tracking method of the exemplary embodiments of the present disclosure will be explained and described in detail with reference to the accompanying drawings.

[0116] First, the nouns involved in the exemplary embodiments of the present disclosure will be explained and described.

[0117] Topic: A topic is the most basic concept in topic detection and tracking. A topic refers to an event that is caused by a certain reason, occurs at a specific time point or time period, within a certain geographical range, and may lead to certain inevitable results; a topic not only includes subsequent activities caused or led by the initial event, but also includes other events directly related to it.

[0118] Topic Detection and Tracking: Topic detection discovers unknown new topics; topic tracking links subsequent related reports with the current topic based on multiple reports on a known topic. Topic detection and tracking technology automatically identifies new topics and tracks known topics in news media information streams. It can mine events from a large amount of breaking news, expand news in real time, and enhance the perception of news hotspots, thereby improving user experience. At the same time, clustering algorithms are used to control news diversity and solve the problem of content homogeneity.

[0119] News features: mining useful information from unstructured text data, accurately summarizing and representing text content, and achieving the ability to distinguish target text from other texts; news features can include keywords and interest tags. Among them, keywords refer to tags directly obtained from news texts using extraction technologies such as textRank and neural networks; interest tags refer to tags that can be abstracted from news texts by describing user interests through user browsing behavior; named entities refer to proper names and meaningful quantitative phrases that appear in texts, and are classified and distinguished.

[0120] Community detection algorithm: A community in a network is composed of a group of nodes that are highly connected to each other, rather than having relatively random and scattered relationships like other nodes in the network; the key to community detection algorithms can be used to extract useful information from the network. Among them, the Louvain algorithm is a community discovery algorithm based on graph networks. The optimization goal is to maximize the modularity of the entire data. The Louvain algorithm is highly efficient and does not require manual specification of the number of communities. It can automatically mine the community with the highest modularity.

[0121] Word2vec: Word2vec is a word embedding method. The model can obtain the distributed vector representation of words, so that words in the same context tend to have the same semantic representation. Among them, Word2vec predicts the current word based on the words in the context window of the word, or predicts the words in the context window based on the current word. The two ideas correspond to the CBOW and skip-gram models respectively.

[0122] Sentence-bert: Bert is a deep bidirectional, unsupervised language representation model that only uses a plain text corpus for pre-training. The structure of the Bert model makes it unsuitable for semantic similarity search or unsupervised clustering tasks. Sentence-bert uses the framework of the twin network model to input two different sentences into two Bert models to obtain the representation of the two sentences. After large-scale data training, the model can calculate the similarity between the two sentences.

[0123] Second, the invention objectives of the exemplary embodiments of the present disclosure are explained and described.

[0124] Specifically, the exemplary embodiments of the present disclosure improve the accuracy of topic aggregation, enrich the diversity of news recommendations, and enhance the continuity of the same topic in the recommendations by automatically mining topics in news media, identifying new topics and tracking existing topics in the news media information flow, thereby enhancing the user's perception and attention to news reports on the same topic. Moreover, through the community discovery algorithm, a two-layer hierarchical structure is constructed by using the interest points and keywords of articles across article categories. Further, in combination with the online clustering method, news is mounted to a specific topic tree by using a multi-dimensional similarity measurement method, and a fault tolerance mechanism and timeliness control are introduced.

[0125] Meanwhile, the exemplary embodiments of the present disclosure define an event as the finest-grained concept, representing a set of one or more documents with similar information; define a topic as a set of multiple logically related events, describing a series of events related to a specific place and specific people during a specific time period; and define a general concept topic as the top-level concept, representing a set of multiple sub-topics related to a theme. Specifically, as shown in Figure 2 the general concept topic may include a first-layer general concept topic and a second-layer general concept topic. The first-layer general concept topic may be, for example, entertainment news, and the second-layer general concept topic may be, for example, celebrity marriages and relationships. That is to say, the first-layer general concept topic may include multiple second-layer general concept topics related to a theme, and the second-layer general concept topic may contain multiple sub-topics related to a theme (such as sub-topic 1, sub-topic 2,..., sub-topic n, etc.). For example, sub-topic 1 may be: "XX Zhou and XX Luo break up", and sub-topic 2 may be: "XX Qu denies relationship", etc. Each sub-topic contains multiple related events, and each news article can be regarded as an event. To solve the problem of homogenization in recommendations, this solution uses a clustering algorithm to divide events into event 1, event 2,..., event n, etc.

[0126] Furthermore, the generation method of the preset two-layer general concept topic tree involved in the exemplary embodiments of the present disclosure is explained and described. Specifically, as shown in Figure 3 the generation method of the preset two-layer general concept topic tree may include the following steps:

[0127] Step S310, obtain historical news events, and use a preset interest tag extraction model to extract second interest tags of the historical news events.

[0128] In this exemplary embodiment, first, historical news events can be obtained. The historical news events can be, for example, news events in the past two weeks, and the number can reach more than 2 million. Then, based on a preset interest tag extraction model, second interest tags of each historical news event are extracted (the "second" here only serves to distinguish from the first interest tag of the current news event and has no other special meaning). It should be supplemented here that the preset interest tag extraction model can be a deep neural network model or other network models, and this example does not make special restrictions on this.

[0129] Step S320: According to the second interest tags and the first co-occurrence relationship between the second interest tags, construct a first network graph, and segment the first network graph to obtain multiple first sub-graph segmentation results; where each first sub-graph segmentation result includes multiple second interest tags.

[0130] In this exemplary embodiment, first, a first network graph is constructed according to the second interest tags and the first co-occurrence relationship between the second interest tags. Specifically, it may include: First, abstract the second interest tags and the first co-occurrence relationship between the second interest tags to obtain vertices and edges; second, calculate the first co-occurrence times of the second interest tags in the same historical news event, and abstract the first co-occurrence times to obtain the weights of the edges; finally, based on the vertices, edges, and edge weights, construct a first network graph.

[0131] Specifically, taking the second interest tags as nodes and the co-occurrence relationship between the second interest tags as edges, construct a first network graph G poi , where if two second interest tags are marked in the same article (the same historical news event), there is an edge between the two second interest tags, and the number of times the two second interest tags appear in the same historical news event is accumulated to obtain the weight of the edge; at the same time, control rules can also be introduced, such as when the co-occurrence frequency is not lower than a certain threshold and the conditional probability is not lower than a certain threshold, it is considered that there is an edge between the two second interest tags; where the calculation method of the conditional probability is: the ratio of the number of times two second interest tags co-occur in the same historical news event to the square root of the number of times each second interest tag appears in the historical news event.

[0132] Secondly, segment the first network graph to obtain multiple first sub-graph segmentation results. Specifically, referring to Figure 4 shown, it may include the following steps:

[0133] S1: Based on a preset community discovery algorithm, establish a split / coalescence tree corresponding to the vertices included in the first network graph;

[0134] S2. Calculate the target weight of each sub - graph segmentation result according to the degree of each vertex and the number of the expected first - sub - graph segmentation results, where the degree of a vertex is the sum of the weights of the edges connected to the vertex.

[0135] S3. Traverse the split / coalescence tree based on a preset order, and determine whether the degree of the vertices under the tree nodes included in the split / coalescence tree is greater than the target weight.

[0136] S4. If the degree of the vertices under any tree node is greater than the target weight, further search the branches of this tree node; otherwise, stop the search and use the vertex set corresponding to this tree node as a first - sub - graph segmentation result.

[0137] S5. Repeat steps S4 and S5 until the total weight of the vertices under all tree nodes is less than or equal to the target weight.

[0138] Specifically, based on steps S1 - S5, the community discovery algorithm can be used to mine sub - graphs in the graph, and the first network graph G poi is segmented into several communities (first - sub - graph segmentation results):

[0139] C poi ={C p1 , C p2 ,..., C pc};

[0140] Among them, each C pi contains multiple interest points, and an interest point may also appear in multiple C pi . Using this method, the first - layer general - concept topics are obtained, and the label features of this layer are represented by the keywords in the set, and the semantic features are the mean vectors represented by the keyword vectors.

[0141] Step S330. Construct a keyword set according to the second keywords of the historical news events included in each of the first - sub - graph segmentation results, and construct a second network graph according to the second keywords included in the keyword set and the second co - occurrence relationship between the second keywords.

[0142] Specifically, the construction process of the second network graph may include: abstracting the second keywords and the second co - occurrence relationship between the second keywords to obtain vertices and edges; calculating the second co - occurrence times of the second keywords in the same historical news event, and abstracting the second co - occurrence times to obtain the weights of the edges; constructing the second network graph based on the vertices, edges, and the weights of the edges. That is, each first - sub - graph segmentation result C pi can be further subdivided, and the second - layer general - concept topics are constructed using the keywords of the historical news events, and the sub - graph C piThe keywords corresponding to the articles of the points of interest form a keyword set C pi K, and a second network graph is constructed based on the co-occurrence relationship between the keywords

[0143] Step S340, segment the second network graph to obtain multiple second sub-graph segmentation results, and generate the preset two-layer general concept topic tree according to the first sub-graph segmentation result and the second sub-graph segmentation result

[0144] Specifically, after obtaining the second network graph, the community discovery algorithm can be used to identify sub-graphs and obtain community C pi The corresponding sub-community (second sub-graph segmentation result) C pi K; where

[0145] C pi K = {C pi K1, C pi K2,..., C pi K k};

[0146] Each C pi The K label feature is characterized by the keywords in the secondary community, and the semantic feature is the average vector of the keywords. Thus, the preset two-layer general concept topic can be obtained according to the first sub-graph segmentation result and the second sub-graph segmentation result, representing topic characterizations of different granularities. It should be supplemented here that the specific segmentation principle of the second network graph is similar to that of the first network graph and will not be elaborated here

[0147] Furthermore, in order to improve the accuracy of the above first sub-graph segmentation result and the second sub-graph segmentation result, that is, to improve the accuracy of the preset two-layer general concept topic tree, the news event tracking method further includes: calculating the first modularity of each of the first sub-graph segmentation results, and evaluating the first sub-graph segmentation result according to the first modularity

[0148] Specifically, since the Louvain algorithm is a community discovery algorithm based on modularity, which can discover hierarchical community structures, and the optimization goal is to maximize the modularity of the entire community network, and at the same time, the near-linear complexity of this algorithm can ensure the program operation efficiency. Modularity is a measurement method for evaluating the quality of a community network partition, and its physical meaning is the difference between the number of edges connecting nodes in the community and the number of edges in the random case, and its value range is [-1 / 2, 1). The definition is as follows: The objective function Modularity function can be expressed as shown below

[0149]

[0150]

[0151] Among them, Aij is the adjacency matrix, representing the weight of the edge between node i and node j. When the network is an unweighted graph, the weight of the edge is 1; k i is the degree of node i, representing the sum of the weights of the edges connected to node i, k j is the degree of node j, representing the sum of the weights of the edges connected to node j, c i and c j respectively represent the communities (subgraph segmentation results) to which node i and node j belong. represents the sum of the weights of all edges. The Louvain community discovery results include multiple clustering graphs, which increase in ascending order of the Modularity value; therefore, the final result with the largest Modularity value can be directly selected.

[0152] Here, it needs to be further supplemented and explained that considering the large number of daily increasing news events, in order to reduce the time of subsequent similar calculations, this solution first constructs a two-layer general concept topic tree using interest tags and keywords in sequence. Compared with directly using classification to detect and track topics within categories, the points of interest can cross categories, associate news of different categories in another dimension, and further improve the mounting efficiency and accuracy of the current news events, so that the corresponding news events can be accurately tracked according to interest tags and keywords, thereby enhancing the user experience.

[0153] Next, the Figures 2 - 4 will be combined with Figure 1 to further explain and illustrate the news event tracking method shown in Figure 1 In a news event tracking method provided in

[0154] In step S110, extract the first interest tag of the current news event, and match the second interest tag corresponding to the first interest tag in the topic hierarchy of the preset two-layer general concept topic tree; wherein, the preset two-layer general concept topic tree is generated according to historical news events.

[0155] In this exemplary embodiment, first, based on the aforementioned preset interest tag extraction model, extract the first interest tag of the current news event; wherein, the current news event is the news event that is updated in real time; second, after obtaining the first interest tag, match the second interest tag corresponding to it in the two-layer general concept topic tree, and then match the general concept topic of the first layer. Here, it needs to be supplemented and explained that since the categories of general concept topics can be enumerated, such as star marriages, current affairs news, agriculture, film and television, social news, etc., there will be no situation where the first interest tag fails to match.

[0156] In step S120, in the keyword level under the second interest tag of the preset two-layer general concept topic tree, a second keyword corresponding to the first keyword of the current news event is matched.

[0157] Specifically, after a corresponding second interest tag is matched, a second keyword corresponding to the first keyword can be matched in the keyword level under the second interest tag (such as star marriage and love); among them, the interest tag can include the first-layer general concept topic and the second-layer general concept topic, that is, the interest tag can be star entertainment → star marriage and love, etc.

[0158] In step S130, according to the first-level position of the second interest tag in the topic level and the second-level position of the second keyword in the keyword level, the mounting position of the current news event in the preset two-layer general concept topic tree is determined.

[0159] In this exemplary embodiment, first, it is determined whether there is a second keyword corresponding to the first keyword in the keyword level; secondly, when it is determined that there is a second keyword corresponding to the first keyword in the keyword level, according to the first-level position of the second interest tag in the topic level and the second-level position of the second keyword in the keyword level, the mounting position of the current news event in the preset two-layer general concept topic tree is determined; further, when it is determined that there is no second keyword corresponding to the first keyword in the keyword level, the first keyword is added to the keyword level; according to the first-level position of the second interest tag in the topic level and the second-level position of the first keyword in the keyword level, the mounting position of the current news event in the preset two-layer general concept topic tree is determined.

[0160] Specifically, if there is a second keyword corresponding to the first keyword in the keyword level, the current news event can be sequentially mounted to the general concept topic tree based on the mounting position; if it is an unknown new topic (there is no second keyword corresponding to the first keyword), a new sub-topic is created using the keyword of the current news event, otherwise it belongs to the existing sub-topic tree. It should be further supplemented here that the keyword level described here refers to the sub-topic event level, that is, in the matching process, the first-layer general concept topic and the second-layer general concept topic are first determined, and then a second keyword corresponding to the current news event is matched under the second-layer general concept topic. The second keyword can be, for example, the star name under star marriage and love, such as Yang X, Li XX, etc.

[0161] In step S140, according to the mounting position, the current news event is mounted to the preset two-layer general concept topic tree to track the current news event according to the second interest tag and / or the second keyword.

[0162] Specifically, the current news event obtains the corresponding mounting position through layer-by-layer matching, and then it can be mounted to the second-level general concept topic tree Tree to which it belongs according to the mounting position. ij , so that the current news event can be tracked according to the interest tag and / or keyword. It should be supplemented here that since there may be a situation where the first keyword does not match the second keyword, the expression "or" is used; that is, the user can directly find the corresponding news event under the corresponding interest tag, or find the corresponding news event under the keyword of the interest tag. This example has no special restrictions on this.

[0163] The following takes the current news event that Yang X opened an entertainment studio as an example to explain and illustrate the specific mounting process of the current news event. Specifically, first, determine the first-level general concept topic "Entertainment News", then the second-level general concept topic "Star Entertainment", and then determine the keyword "Yang" under the second-level general concept topic, and then mount the current news event to the sub-event corresponding to "Yang"; of course, if there is already a sub-event corresponding to the entertainment studio in the sub-event, it can be merged; if not, the sub-event can be mounted under the keyword with the entertainment studio as the keyword.

[0164] Furthermore, in order to reduce the redundancy of news events, as shown in Figure 5 , the news event tracking method may further include the following steps:

[0165] Step S510, calculate the similarity between the current news event and the historical news event corresponding to the second keyword.

[0166] In this example embodiment, first, the current news event is input into the first Sentence-Bert model to obtain a first vector representation, and the historical news event corresponding to the second keyword is input into the second Sentence-Bert model to obtain a second vector representation; secondly, calculate the cosine similarity between the first vector representation and the second vector representation to obtain the similarity between the current news event and the historical news event corresponding to the second keyword.

[0167] Specifically, the first Sentence-Bert model and the second Sentence-Bert model can be trained through a large-scale corpus. In this example embodiment, as shown inFigure 6 As shown in the figure, the current news event and the historical news event corresponding to the second keyword are respectively input into the first Sentence-Bert model 602 and the second Sentence-Bert model 603 through the input layer 601, so as to obtain the first vector representation and the second vector representation; then, the cosine similarity between the first vector representation and the second vector representation is calculated through the similarity calculation layer 604, and then the similarity value is output through the output layer 605; wherein, the parameters of the first Sentence-Bert model and the second Sentence-Bert model are shared, and the larger the similarity value is, the more similar the news events are, otherwise they are dissimilar.

[0168] It should be noted here that in addition to the conventional calculation of the cosine similarity of word features (word overlap, named entity overlap, location overlap), the semantic features and the word2vec word embedding method can be used to obtain the distributed vector representation of words, and the semantic representation of sentences can be represented by the vector representation of words. Similarly, after the sentence is input into large-scale pre-trained models such as BERT, the vector of the last token of the model can be used as the vector representation of the sentence. However, the semantic representations generated by the above two methods have little association with the target task, so the effect is not obvious when used in specific tasks. Therefore, the first Sentence-Bert model and the second Sentence-Bert model are selected in this exemplary embodiment.

[0169] Step S520, when it is determined that the similarity is greater than the first preset threshold, merge the current news event and the historical news event corresponding to the second keyword, and update the historical news event corresponding to the second keyword by using the merged news event.

[0170] Specifically, if the similarity is greater than the first preset threshold (for example, 0.7 or 0.8, and of course it can also be other values, and this example does not make special restrictions on this), then the current news event and the historical news event are merged and updated; of course, if the similarity is less than or equal to the first preset threshold, it is determined that the current news event is an unknown new topic, and then a new sub-topic is created under the Tree ij The feature representation of the sub-topic is the feature of this article, such as keywords, semantics, named entities, etc.

[0171] Furthermore, in order to avoid the problem of low accuracy of the mounting position caused by the overly broad topic bucket, as shown in the reference Figure 7 The news event tracking method may further include the following steps:

[0172] Step S710: Calculate the number of historical news events and / or current news events included under keywords of different categories at the keyword hierarchy level.

[0173] Step S720: When it is determined that the number of historical news events and / or current news events included under the keywords of any category is greater than the second preset threshold, split the historical news events and / or current news events included under the keywords of this category, and update the keywords of this category according to the split results.

[0174] Hereinafter, Steps S710 - S720 will be explained and described. Specifically, the topic model timing update module involves error correction and timeliness control; after multiple matches and mountings, the features of the topic tree may be too scattered, tending to aggregate several topics together. At this time, the formed topic is too large, not concentrated enough, and the interest distribution is too wide. Therefore, error correction is performed on sub-topics and events, and clustering division is performed on sub-topics and events with a large number of included news. Limited by storage space and to reduce the computational complexity, this solution regularly deletes expired articles and updates the features of sub-topics and events. This solution defines that news becomes invalid after two weeks.

[0175] Specifically, for error correction of sub-topics and events, it means merging and splitting the sub-topics and events mounted on the third and fourth layers respectively; due to the uncontrollability of streaming clustering, after a long time, relatively broad topic buckets will appear. It is agreed that further splitting is required after the aggregated number of articles exceeds a certain amount. In this exemplary embodiment, the DBSCAN algorithm is used to split within a relatively large sub-topic. DBSCAN is a density-based clustering method that divides closely connected data into the same cluster and can cluster dense data sets of any shape with less interference from noise points and without setting the number of clusters.

[0176] It can be learned so far that the news event tracking method provided by the exemplary embodiments of the present disclosure may include three parts: generating an offline two-layer general concept topic tree, mounting the current news event, and periodically updating the two-layer general concept topic tree; among them, the periodic update of the two-layer general concept topic tree may include updating the second-layer general concept topics and / or sub-topics and / or events. For example, new general concept topics may be added to the second-layer general concept topics, new sub-topics may be added to the sub-topics, and new events may be added to the events, etc.; and, the two-layer general concept topic tree constructed by the exemplary embodiments of the present disclosure can accurately mount the current news event, thereby effectively improving the diversity of recommendations, enhancing the immersive reading experience, increasing the click-through rate and the user browsing duration; at the same time, since the current news event can be processed in real time, and the current news event can be mounted to the corresponding sub-topic by combining a multi-dimensional similarity measurement mechanism, and periodically updated dynamically and corrected for errors, the accuracy of news event mounting is further improved.

[0177] Exemplary Apparatus

[0178] The present disclosure also provides a news event tracking device. Refer to Figure 8 As shown, the news event tracking device may include a first matching module 810, a second matching module 820, a mounting position determination module 830, and an event tracking module 840. Among them:

[0179] The first matching module 810 may be configured to extract the first interest label of the current news event, and match the second interest label corresponding to the first interest label in the topic level of the preset two-layer general concept topic tree; wherein, the preset two-layer general concept topic tree is generated according to historical news events;

[0180] The second matching module 820 may be configured to match the second keyword corresponding to the first keyword of the current news event in the keyword level under the second interest label;

[0181] The mounting position determination module 830 may be configured to determine the mounting position of the current news event in the preset two-layer general concept topic tree according to the first-level position of the second interest label in the topic level and the second-level position of the second keyword in the keyword level;

[0182] The event tracking module 840 may be configured to mount the current news event to the preset two-layer general concept topic tree according to the mounting position, so as to track the current news event according to the second interest label and / or the second keyword.

[0183] In an exemplary embodiment of the present disclosure, determining the mounting position of the current news event in the preset two - layer general concept topic tree according to the first - layer position of the second interest tag in the topic hierarchy and the second - layer position of the second keyword in the keyword hierarchy includes:

[0184] When it is determined that there is a second keyword corresponding to the first keyword in the keyword hierarchy, determine the mounting position of the current news event in the preset two - layer general concept topic tree according to the first - layer position of the second interest tag in the topic hierarchy and the second - layer position of the second keyword in the keyword hierarchy.

[0185] In an exemplary embodiment of the present disclosure, the tracking device for news events further includes:

[0186] A first adding module, configured to add the first keyword to the keyword hierarchy when it is determined that there is no second keyword corresponding to the first keyword in the keyword hierarchy;

[0187] A first position determining module, configured to determine the mounting position of the current news event in the preset two - layer general concept topic tree according to the first - layer position of the second interest tag in the topic hierarchy and the second - layer position of the first keyword in the keyword hierarchy.

[0188] In an exemplary embodiment of the present disclosure, the tracking device for news events further includes:

[0189] A first similarity calculation module, configured to calculate the similarity between the current news event and the historical news event corresponding to the second keyword;

[0190] An event merging module, configured to merge the current news event and the historical news event corresponding to the second keyword when it is determined that the similarity is greater than a first preset threshold, and update the historical news event corresponding to the second keyword with the merged news event.

[0191] In an exemplary embodiment of the present disclosure, calculating the similarity between the current news event and the historical news event corresponding to the second keyword includes:

[0192] Input the current news event into a first Sentence - Bert model to obtain a first vector representation, and input the historical news event corresponding to the second keyword into a second Sentence - Bert model to obtain a second vector representation;

[0193] Calculate the cosine similarity between the first vector representation and the second vector representation to obtain the similarity between the current news event and the historical news event corresponding to the second keyword.

[0194] In an exemplary embodiment of the present disclosure, the tracking device for news events further includes:

[0195] An interest tag extraction module, configured to obtain historical news events and use a preset interest tag extraction model to extract second interest tags of the historical news events;

[0196] A first graph segmentation module, configured to construct a first network graph according to the second interest tags and a first co-occurrence relationship between the second interest tags, and segment the first network graph to obtain a plurality of first sub-graph segmentation results; wherein, in each of the first sub-graph segmentation results, a plurality of second interest tags are included;

[0197] A network graph construction module, configured to construct a keyword set according to second keywords of historical news events included in each of the first sub-graph segmentation results, and construct a second network graph according to the second keywords included in the keyword set and a second co-occurrence relationship between the second keywords;

[0198] A second graph segmentation module, configured to segment the second network graph to obtain a plurality of second sub-graph segmentation results, and generate the preset two-layer general concept topic tree according to the first sub-graph segmentation results and the second sub-graph segmentation results.

[0199] In an exemplary embodiment of the present disclosure, constructing a first network graph according to the second interest tags and a first co-occurrence relationship between the second interest tags includes:

[0200] Perform an abstraction process on the second interest tags and the first co-occurrence relationship between the second interest tags to obtain vertices and edges;

[0201] Calculate the first co-occurrence times of the second interest tags in the same historical news event, and perform an abstraction process on the first co-occurrence times to obtain the weights of the edges;

[0202] Construct a first network graph based on the vertices, edges, and weights of the edges.

[0203] In an exemplary embodiment of the present disclosure, segmenting the first network graph to obtain a plurality of first sub-graph segmentation results includes:

[0204] S1, establish a split / coalescence tree corresponding to the vertices included in the first network graph based on a preset community discovery algorithm;

[0205] S2. Calculate the target weight of each sub - graph segmentation result according to the degree of each vertex and the number of the expected first sub - graph segmentation results, where the degree of the vertex is the sum of the weights of the edges connected to the vertex.

[0206] S3. Traverse the split / merge tree based on a preset order, and determine whether the degree of the vertices under the tree nodes included in the split / merge tree is greater than the target weight.

[0207] S4. If the degree of the vertices under any tree node is greater than the target weight, further search the branches of this tree node; otherwise, stop the search and use the vertex set corresponding to this tree node as a first sub - graph segmentation result.

[0208] S5. Repeat steps S4 and S5 until the total weight of the vertices under all tree nodes is less than or equal to the target weight.

[0209] In an exemplary embodiment of the present disclosure, the news event tracking device further includes:

[0210] An event quantity calculation module, configured to calculate the quantity of historical news events and / or current news events included in different keywords of each category at the keyword level.

[0211] An event splitting module, configured to split the historical news events and / or current news events included in the keywords of any category when it is determined that the quantity of the historical news events and / or current news events included in the keywords of this category is greater than a second preset threshold, and update the keywords of this category according to the splitting result.

[0212] Exemplary Storage Medium

[0213] After introducing the news event tracking method and the news event tracking device of the exemplary embodiments of the present disclosure, next, refer to Figure 9 to describe the storage medium of the exemplary embodiments of the present disclosure.

[0214] Refer to Figure 9 As shown, a program product 900 for implementing the above - mentioned method according to an embodiment of the present disclosure is described. It can use a portable compact disc read - only memory (CD - ROM) and include program code, and can run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto.

[0215] The program product may employ any combination of one or more readable media. The readable media may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples of the readable storage medium (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0216] A computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which the readable program code is carried. Such a propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. The readable signal medium may also be any readable medium other than the readable storage medium.

[0217] The program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages such as Java, C++, etc., and also including conventional procedural programming languages such as the "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN).

[0218] Exemplary Electronic Device

[0219] After introducing the storage medium of the exemplary embodiments of the present disclosure, next, reference is made to Figure 10 the electronic device of the exemplary embodiments of the present disclosure is described.

[0220] Figure 10 The displayed electronic device 1000 is merely an example and should not impose any limitation on the functions and the scope of use of the embodiments of the present disclosure.

[0221] As Figure 10As shown, the electronic device 1000 is presented in the form of a general computing device. The components of the electronic device 1000 may include, but are not limited to: at least one of the above-mentioned processing units 1010, at least one of the above-mentioned storage units 1020, a bus 1030 connecting different system components (including the storage unit 1020 and the processing unit 1010), and a display unit 1040.

[0222] Among them, the storage unit 1020 stores program codes, and the program codes can be executed by the processing unit 1010, so that the processing unit 1010 executes the steps according to various exemplary embodiments of the present disclosure described in the "Exemplary Method" section of the present specification above. For example, the processing unit 1010 can execute as Figure 1 shown in: Step S110. Extract the first interest tag of the current news event, and match the second interest tag corresponding to the first interest tag in the topic level of the preset two-layer general concept topic tree; wherein, the preset two-layer general concept topic tree is generated according to historical news events; Step S120. In the keyword level under the second interest tag of the preset two-layer general concept topic tree, match the second keyword corresponding to the first keyword of the current news event; Step S130. Determine the mounting position of the current news event in the preset two-layer general concept topic tree according to the first-level position of the second interest tag in the topic level and the second-level position of the second keyword in the keyword level; Step S140: Mount the current news event to the preset two-layer general concept topic tree according to the mounting position, so that the current news event can be tracked according to the second interest tag and / or the second keyword.

[0223] The storage unit 1020 may include a volatile storage unit, such as a random access storage unit (RAM) 10201 and / or a cache storage unit 10202, and may further include a read-only storage unit (ROM) 10203.

[0224] The storage unit 1020 may further include a program / utilities 10204 having a set (at least one) of program modules 10205. Such program modules 10205 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include the implementation of a network environment.

[0225] The bus 1030 may include a data bus, an address bus, and a control bus.

[0226] The electronic device 1000 can also communicate with one or more external devices 1100 (such as a keyboard, a pointing device, a Bluetooth device, etc.) through the input / output (I / O) interface 1050. Moreover, the electronic device 1000 can also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through the network adapter 1060. As shown in the figure, the network adapter 1060 communicates with other modules of the electronic device 1000 through the bus 1030. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in combination with the electronic device 1000, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0227] It should be noted that although several modules or sub-modules of the news event tracking device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more of the above-described units / modules can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0228] It should be noted that although several units / modules or sub-units / modules of the device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more of the above-described units / modules can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0229] In addition, although the operations of the method of the present disclosure are described in a specific order in the drawings, this does not require or imply that these operations must be performed in that specific order, or that all of the shown operations must be performed to achieve the desired result. Additionally or alternatively, some steps can be omitted, multiple steps can be combined into one step for execution, and / or one step can be decomposed into multiple steps for execution.

[0230] Although the spirit and principles of the present disclosure have been described with reference to several specific embodiments, it should be understood that the present disclosure is not limited to the disclosed specific embodiments, and the division of each aspect does not mean that the features in these aspects cannot be combined for benefit. This division is only for the convenience of expression. The present disclosure aims to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A method for tracking news events, comprising: Extract the first interest tag of the current news event and match the second interest tag corresponding to the first interest tag in the topic levels of the preset two-layer general concept topic tree; wherein, the preset two-layer general concept topic tree is generated based on historical news events; Match the second keyword corresponding to the first keyword of the current news event in the keyword level under the second interest tag of the preset two-layer general concept topic tree; When it is determined that there is a second keyword corresponding to the first keyword in the keyword level, determine the mounting position of the current news event in the preset two-layer general concept topic tree according to the first-level position of the second interest tag in the topic level and the second-level position of the second keyword in the keyword level; when it is determined that there is no second keyword corresponding to the first keyword in the keyword level, add the first keyword to the keyword level; determine the mounting position of the current news event in the preset two-layer general concept topic tree according to the first-level position of the second interest tag in the topic level and the second-level position of the first keyword in the keyword level; Mount the current news event to the preset two-layer general concept topic tree according to the mounting position, so as to track the current news event according to the second interest tag and / or the second keyword.

2. The method for tracking news events according to claim 1, wherein, After determining the mounting position of the current news event in the preset two-layer general concept topic tree, the news event tracking method further includes: Calculate the similarity between the current news event and the historical news events corresponding to the second keyword; When it is determined that the similarity is greater than the first preset threshold, merge the current news event and the historical news events corresponding to the second keyword, and update the historical news events corresponding to the second keyword with the merged news event.

3. The method for tracking news events according to claim 2, wherein, Calculating the similarity between the current news event and the historical news events corresponding to the second keyword includes: Input the current news event into the first Sentence-Bert model to obtain a first vector representation, and input the historical news events corresponding to the second keyword into the second Sentence-Bert model to obtain a second vector representation; Calculate the cosine similarity between the first vector representation and the second vector representation to obtain the similarity between the current news event and the historical news events corresponding to the second keyword.

4. The method for tracking news events according to claim 1, wherein, The news event tracking method further includes: Obtain historical news events, and use a preset interest tag extraction model to extract the second interest tags of the historical news events; Construct a first network graph according to the second interest tags and the first co-occurrence relationship between the second interest tags, and segment the first network graph to obtain a plurality of first sub-graph segmentation results; wherein, each of the first sub-graph segmentation results includes a plurality of second interest tags; the first co-occurrence relationship means that two or more of the second interest tags appear in the same historical news event; Construct a keyword set according to the second keywords of the historical news events included in each of the first sub-graph segmentation results, and construct a second network graph according to the second keywords included in the keyword set and the second co-occurrence relationship between the second keywords; Segment the second network graph to obtain a plurality of second sub-graph segmentation results, and generate the preset two-layer general concept topic tree according to the first sub-graph segmentation results and the second sub-graph segmentation results.

5. The method for tracking news events according to claim 4, wherein, Constructing a first network graph according to the second interest tags and the first co-occurrence relationship between the second interest tags includes: Perform an abstraction process on the second interest tags and the first co-occurrence relationship between the second interest tags to obtain vertices and edges; Calculate the first co-occurrence times of the second interest tags in the same historical news event, and perform an abstraction process on the first co-occurrence times to obtain the weights of the edges; Construct a first network graph based on the vertices, edges and the weights of the edges.

6. The method for tracking news events according to claim 5, wherein, Segmenting the first network graph to obtain a plurality of first sub-graph segmentation results includes: S1, establish a split / coalescence tree corresponding to the vertices included in the first network graph based on a preset community discovery algorithm; S2, calculate the target weights of the sub-graph segmentation results according to the degrees of the vertices and the number of the first sub-graph segmentation results expected to be obtained; wherein, the degree of a vertex is the sum of the weights of the edges connected to the vertex; S3, traverse the split / coalescence tree in a preset order, and judge whether the degree of the vertices under the tree nodes included in the split / coalescence tree is greater than the target weight; S4, if the degree of the vertices under any tree node is greater than the target weight, further search the branches of the tree node; otherwise, stop the search, and use the vertex set corresponding to the tree node as a first sub-graph segmentation result; S5, repeat steps S4 and S5 until the total weights of the vertices under all the tree nodes are less than or equal to the target weight.

7. The method for tracking news events according to claim 6, wherein, The news event tracking method further includes: Calculate the first modularity of each of the first sub-graph segmentation results, and evaluate the first sub-graph segmentation results according to the first modularity.

8. The method for tracking news events according to claim 1, wherein, The news event tracking method further includes: Calculate the number of historical news events and / or current news events included under keywords of different categories at the keyword level; When it is determined that the number of historical news events and / or current news events included under the keywords of any category is greater than a second preset threshold, split the historical news events and / or current news events included under the keywords of this category, and update the keywords of this category according to the split results.

9. A tracking device for news events, comprising: A first matching module, configured to extract a first interest tag of a current news event, and match a second interest tag corresponding to the first interest tag in a topic level of a preset two-layer general concept topic tree; wherein, the preset two-layer general concept topic tree is generated according to historical news events; A second matching module, configured to match a second keyword corresponding to a first keyword of the current news event in a keyword level under the second interest tag; A mounting position determining module, configured to, when determining that there is a second keyword corresponding to the first keyword in the keyword level, determine a mounting position of the current news event in the preset two-layer general concept topic tree according to a first-level position of the second interest tag in the topic level and a second-level position of the second keyword in the keyword level; when determining that there is no second keyword corresponding to the first keyword in the keyword level, add the first keyword to the keyword level; and determine a mounting position of the current news event in the preset two-layer general concept topic tree according to a first-level position of the second interest tag in the topic level and a second-level position of the first keyword in the keyword level; An event tracking module, configured to mount the current news event to the preset two-layer general concept topic tree according to the mounting position, so as to track the current news event according to the second interest tag and / or the second keyword.

10. The tracking device for news events according to claim 9, wherein, The tracking device for news events further includes: A first similarity calculation module, configured to calculate a similarity between the current news event and a historical news event corresponding to the second keyword; An event merging module, configured to, when determining that the similarity is greater than a first preset threshold, merge the current news event and the historical news event corresponding to the second keyword, and update the historical news event corresponding to the second keyword by using the merged news event.

11. The tracking device for news events according to claim 10, wherein, Calculating the similarity between the current news event and the historical news event corresponding to the second keyword includes: Inputting the current news event into a first Sentence-Bert model to obtain a first vector representation, and inputting the historical news event corresponding to the second keyword into a second Sentence-Bert model to obtain a second vector representation; Calculating a cosine similarity between the first vector representation and the second vector representation to obtain the similarity between the current news event and the historical news event corresponding to the second keyword.

12. The tracking device for news events according to claim 9, wherein, The tracking device for news events further includes: An interest tag extraction module, configured to obtain historical news events, and extract second interest tags of the historical news events by using a preset interest tag extraction model; The first graph segmentation module is used to construct a first network graph according to the second interest tags and the first co-occurrence relationship between the second interest tags, and segment the first network graph to obtain a plurality of first sub-graph segmentation results; wherein, in each of the first sub-graph segmentation results, a plurality of second interest tags are included; the first co-occurrence relationship means that two or more of the second interest tags appear in the same historical news event; The network graph construction module is used to construct a keyword set according to the second keywords of the historical news events included in each of the first sub-graph segmentation results, and construct a second network graph according to the second keywords included in the keyword set and the second co-occurrence relationship between the second keywords; The second graph segmentation module is used to segment the second network graph to obtain a plurality of second sub-graph segmentation results, and generate the preset two-layer general concept topic tree according to the first sub-graph segmentation results and the second sub-graph segmentation results.

13. The tracking device for news events according to claim 12, wherein, Constructing a first network graph according to the second interest tags and the first co-occurrence relationship between the second interest tags includes: Performing an abstraction process on the second interest tags and the first co-occurrence relationship between the second interest tags to obtain vertices and edges; Calculating the first co-occurrence times of the second interest tags in the same historical news event, and performing an abstraction process on the first co-occurrence times to obtain the weights of the edges; Constructing a first network graph based on the vertices, edges and the weights of the edges.

14. The tracking device for news events according to claim 13, wherein, Segmenting the first network graph to obtain a plurality of first sub-graph segmentation results includes: S1, establishing a split / coalescence tree corresponding to the vertices included in the first network graph based on a preset community discovery algorithm; S2, calculating the target weights of the sub-graph segmentation results according to the degrees of the vertices and the number of the first sub-graph segmentation results expected to be obtained; wherein, the degree of a vertex is the sum of the weights of the edges connected to the vertex; S3, traversing the split / coalescence tree in a preset order, and judging whether the degree of the vertices under the tree nodes included in the split / coalescence tree is greater than the target weight; S4, if the degree of the vertices under any tree node is greater than the target weight, further searching the branches of the tree node; otherwise, stopping the search, and taking the vertex set corresponding to the tree node as a first sub-graph segmentation result; S5, repeating steps S4 and S5 until the total weights of the vertices under all tree nodes are less than or equal to the target weight.

15. The tracking device for news events according to claim 14, wherein, The news event tracking device further includes: The segmentation result evaluation module is used to calculate the first modularity of each of the first sub-graph segmentation results, and evaluate the first sub-graph segmentation results according to the first modularity.

16. The tracking device for news events according to claim 9, wherein, The news event tracking device further includes: The event quantity calculation module is used to calculate the quantity of the historical news events and / or the current news events included under different keywords of each category at the keyword level; An event splitting module, configured to split historical news events and / or current news events included under keywords of any category when it is determined that the number of historical news events and / or current news events included under the keywords of this category is greater than a second preset threshold, and update the keywords of this category according to the splitting result.

17. A computer-readable storage medium having a computer program stored thereon, where the computer program, when executed by a processor, implements the news event tracking method according to any one of claims 1-8.

18. An electronic device, comprising: A processor; and a memory for storing executable instructions of the processor; wherein, the processor is configured to execute the news event tracking method according to any one of claims 1-8 by executing the executable instructions.

Images