A hotspot information mining method and related device

Abstract

The application discloses a hotspot information mining method and related device, and relates to the field of data processing, comprising: marking a plurality of event data in a demand time period to obtain event labels corresponding to the plurality of event data respectively, the event labels including event types and involved subjects, performing first-level clustering on the plurality of event data according to the event types and event semantic similarity to obtain at least one event cluster, taking an event cluster with an event quantity greater than a first threshold in the at least one event cluster as a hotspot event cluster, performing second-level clustering on the hotspot event cluster according to the involved subjects to obtain at least one event sub-cluster, taking an event sub-cluster with an event quantity greater than a second threshold in the at least one event sub-cluster as a hotspot event sub-cluster, and generating hotspot information according to the hotspot event cluster and the hotspot event sub-cluster. Through the automatic multi-level clustering, the application avoids missing or deviation of the hotspot information, and improves the comprehensiveness and accuracy of the hotspot information.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and related apparatus for mining hotspot information. Background Technology

[0002] The daily influx of citizen requests from various channels constitutes a massive amount of raw data. After being sorted and standardized by staff, this raw data is integrated into an event list, forming an event database reflecting the dynamic information of the city's operations. In order to extract hot topics from the event list containing a massive number of events, staff currently need to analyze the event type of each event in the event list, then group the event data according to event type, and finally analyze and extract hot topics from each group.

[0003] However, manual analysis of event types relies heavily on human experience. Different staff members may label the same event with different event types, leading to discrepancies between manually identified hotspot information and the actual situation. Even if the event type is accurately matched, insufficient analytical depth by staff may result in the omission of hotspot information or discrepancies between the identified hotspot information and the actual situation. This could prevent the timely detection and handling of potential urban operational risks, thus affecting the timeliness of urban operational risk warnings. Summary of the Invention

[0004] In view of the above problems, this application provides a method and related apparatus for hotspot information mining to solve the problem of omission or bias in hotspot information caused by the existing manual analysis of hotspot information. The specific solution is as follows:

[0005] The first aspect of this application provides a method for mining hotspot information, including:

[0006] Multiple event data within the required time period are tagged to obtain event tags corresponding to the multiple event data, and the event tags include event type and involved parties;

[0007] Based on the event type and event semantic similarity, the multiple event data are clustered at the first level to obtain at least one event cluster. The event clusters with more than one threshold of events in the at least one event cluster are regarded as hot event clusters. The event data in each event cluster correspond to the same event type.

[0008] The hot event clusters are clustered at the second level according to the involved entities to obtain at least one event sub-cluster. The event sub-clusters with more than a second threshold in the at least one event sub-cluster are regarded as hot event sub-clusters.

[0009] Hotspot information is generated based on the hotspot event clusters and the hotspot event subclusters.

[0010] In one possible implementation, tagging multiple event data within the required time period includes:

[0011] The configured classification reasoning engine is used to classify the multiple event data into event types, thereby obtaining the event types corresponding to the multiple event data.

[0012] Using the configured first language model, target event elements are extracted from the multiple event data respectively to obtain the target event elements corresponding to the multiple event data, and the target event elements include the involved subject;

[0013] The event data are labeled according to the event type and target event element corresponding to each of the event data.

[0014] In one possible implementation, the training process for the model file used by the classification inference engine includes:

[0015] Using the event classification model pre-built for the classification inference engine, the training event data is classified and predicted to obtain the event type prediction result corresponding to the training event data;

[0016] The second language model is configured to determine whether the training event data belongs to the event type prediction result corresponding to the training event data, and if the determination result is no, a suggested event type is generated;

[0017] The judgment results and / or the suggested event types from the second largest language model are output and displayed to obtain the real event types from human feedback;

[0018] The event classification model is trained using the real event types and the training event data to obtain the trained model file.

[0019] In one possible implementation, the step of performing a first-level clustering of the multiple event data based on the event type and event semantic similarity to obtain at least one event cluster includes:

[0020] The multiple event data are clustered according to the event type to obtain at least one first cluster, and the event data in the first cluster correspond to the same event type;

[0021] The first cluster in which the number of events in at least one first cluster is greater than the third threshold is designated as the second cluster;

[0022] Calculate the semantic similarity of each pair of event data within the second cluster, and cluster the event data within the second cluster based on the calculated semantic similarity to obtain at least one third cluster, which serves as the at least one event cluster.

[0023] In one possible implementation, the target event element further includes a brief description of the event data, and the step of calculating the event semantic similarity pairwise for each event data point within the second cluster includes:

[0024] Using the configured deduplication engine, for every two event data in the second cluster, the event time, event location, and reporter identifier of each of the two event data are determined, and the similarity of the content summary, event time, event location, and reporter identifier of each of the two event data is calculated. Based on the calculated similarity, the event semantic similarity of the two event data is determined.

[0025] In one possible implementation, generating hotspot information based on the hotspot event cluster and the hotspot event sub-clusters includes:

[0026] Determine the location of each event data point within the hotspot event sub-cluster;

[0027] Select a baseline event data point from the hotspot event sub-cluster and place it in the fourth cluster. Then, iterate through the other event data within the hotspot event sub-cluster.

[0028] Determine the distance between the location of the baseline event data and the location of the currently traversed event data. If the distance is less than or equal to a preset distance threshold, then place the currently traversed event data in the fourth cluster.

[0029] At the end of the traversal, it is determined whether the number of events in the fourth cluster is greater than the target number, which is determined based on the number of events in the hot event sub-cluster.

[0030] If not, clear the fourth cluster, and select a new base event data from the hot event sub-cluster and place it in the fourth cluster, then return the other event data within the hot event sub-cluster.

[0031] If so, then hotspot information is generated based on the hotspot event cluster and the fourth cluster.

[0032] In one possible implementation, generating hotspot information based on the hotspot event cluster and the fourth cluster includes:

[0033] Using the configured third language model, generate hot topic titles and descriptions based on the hot topic event clusters and / or the fourth cluster, and determine the summary information of hot topic event handling corresponding to the hot topic event clusters and / or the fourth cluster;

[0034] Determine the influence metric values ​​corresponding to the hot topic event cluster and the fourth cluster, respectively;

[0035] The hot topic information is defined as the hot topic title, hot topic description information, hot topic event handling summary information, and influence measurement value corresponding to the hot topic event cluster and the fourth cluster, respectively.

[0036] One possible implementation also includes:

[0037] Retrieve knowledge fragments related to the hot topic information from a pre-configured knowledge base;

[0038] Based on the knowledge fragments and the hotspot information, generate the hotspot event clusters and / or the corresponding processing measures for the fourth cluster.

[0039] In one possible implementation, the subject involved is referred to as the subject's abbreviation.

[0040] A second aspect of this application provides a computer program product including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the hotspot information mining method described in the first aspect or any implementation thereof.

[0041] A third aspect of this application provides an electronic device, comprising at least one processor and a memory connected to the processor, wherein:

[0042] The memory is used to store computer programs;

[0043] The processor is used to execute the computer program so that the electronic device can implement the hotspot information mining method of the first aspect or any implementation thereof.

[0044] A fourth aspect of this application provides a computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the hotspot information mining method described in the first aspect or any implementation thereof.

[0045] By employing the aforementioned technical solution, the hot topic information mining method provided in this application, in order to mine hot topics within a required time period, labels multiple event data within that time period to obtain event tags corresponding to each event data point. These event tags include event type and involved parties. Next, based on event type and semantic similarity, the multiple event data are clustered at a first level to obtain at least one event cluster. Event clusters with more than a first threshold number of events within these at least one event cluster are designated as hot topic event clusters. Each event cluster within the at least one event cluster corresponds to the same event type. Clustering by event type allows events of the same type to be grouped together, avoiding noise interference from different types of events in the hot topic information mining process. Furthermore, considering that even events belonging to the same event type may reflect different urban operation problems, and a hot topic event only focuses on one type of urban operation problem, semantic similarity clustering is also introduced during event type clustering. This more accurately clusters events reflecting urban operation problems repeatedly mentioned by citizens, improving the accuracy of hot topic information mining and reducing the risk of missing or biased hot topic information.

[0046] Furthermore, considering that a cluster of hot events may be associated with multiple stakeholders, in order to clarify the hot events at the stakeholder level, this application can further perform a second-level clustering of the hot event clusters by stakeholder, obtaining at least one event sub-cluster, and designating the event sub-cluster with the number of events exceeding a second threshold as the hot event sub-cluster. Finally, hot event information is generated based on the hot event clusters and hot event sub-clusters. Therefore, this application considers not only hot events at the event type level but also hot events at the stakeholder level, making the final generated hot event information more comprehensive. By identifying hot events at the stakeholder level, this application can more accurately determine the objects that pose potential risks to social operations, thereby enabling more targeted responses, reducing risk management costs, and improving urban operational efficiency. Attached Figure Description

[0047] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0048] Figure 1 A schematic diagram of a system architecture provided for this application;

[0049] Figure 2 A flowchart illustrating a hotspot information mining method provided in this application;

[0050] Figure 3A schematic diagram illustrating the process of tagging event data provided in this application;

[0051] Figure 4 A schematic diagram of hotspot information corresponding to hotspot event clusters;

[0052] Figure 5 This is a diagram illustrating the hot topics related to the fourth cluster of information concerning the issue of school opening times during hot weather.

[0053] Figure 6 This is a schematic diagram of an analysis report;

[0054] Figure 7 This is a schematic diagram illustrating proposed measures to address the issue of delayed school opening at Jianshe Road Primary School due to high temperatures.

[0055] Figure 8 A schematic diagram of a hotspot information mining device provided in this application;

[0056] Figure 9 This is a schematic diagram of the structure of an electronic device provided in this application. Detailed Implementation

[0057] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0058] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0059] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0060] This application provides a method and related apparatus for hotspot information mining. Optionally, the hotspot information mining method provided in this application can be applied to, for example... Figure 1 The system architecture shown includes a terminal 100 and a server 200. The server 200 may include one or more servers (…).Figure 1 (This example uses a server as an illustration).

[0061] Either terminal 100 or server 200 can be used independently to execute the hotspot information mining method provided in the embodiments of this application. Alternatively, terminal 100 and server 200 can also be used collaboratively to execute the hotspot information mining method provided in the embodiments of this application.

[0062] The following description Figure 1 The product form of the mid-terminal 100;

[0063] The terminal 100 in this application embodiment can be a mobile phone, tablet computer, wearable device, vehicle device, augmented reality (AR) / virtual reality (VR) device, laptop computer, ultra-mobile personal computer (UMPC), netbook, personal digital assistant (PDA), etc., and this application embodiment does not impose any restrictions on it.

[0064] To enable those skilled in the art to better understand this application, the hotspot information mining method of the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0065] Reference Figure 2 , Figure 2 This application provides a flowchart illustrating a hotspot information mining method, as shown in the embodiments below. Figure 2 As shown, this hotspot information mining method may include:

[0066] Step S201: Tag multiple event data within the required time period to obtain event tags corresponding to each event data. The event tags include the event type and the parties involved.

[0067] As described in the background technology, staff can compile citizens' requests into an event list. This event list can include an event code, the time of the incident, and event data. The event code is used to uniquely identify the citizen's request, the time of the incident is the time when the citizen's request occurred, and the event data is a specific description of the citizen's request. For example, a specific event data could be: "The caller reported that around 00:30 on January 9th, there were many mobile vendors occupying the road on xx Road in xx District, causing noise pollution, and the urban management department is not taking action. Please investigate and deal with this matter."

[0068] When a user needs to analyze trending topics, this embodiment can obtain data from multiple events within the desired time period. For example, optionally, an intelligent analysis assistant can be invoked to filter multiple event data within the desired time period from the event list.

[0069] Furthermore, in order to accurately profile and locate each event data within the required time period, this embodiment can label each event data separately to obtain event tags corresponding to each event data.

[0070] Optionally, the event tags include the event type and the parties involved. The parties involved refer to the objects associated with the event data. For example, if the event data is "The caller reported that he worked at the Children's Hospital of xx District from July to September 2023 and has not received his wages to date; the employer in arrears is xx Construction; additional description: xx Construction stated that the money has been paid to xx Construction Labor Service Co., Ltd.", then the parties involved include: xx Construction and xx Construction Labor Service Co., Ltd.

[0071] Step S202: Perform first-level clustering on multiple event data according to event type and event semantic similarity to obtain at least one event cluster. The event cluster with more than one threshold of events in the at least one event cluster is regarded as a hot event cluster. The event data in each event cluster corresponds to the same event type.

[0072] In this embodiment, multiple event data can be clustered according to event type to mine hotspot information (i.e., information describing hot events) at the event type dimension. Considering that a hotspot event only focuses on one type of urban operation problem, even if they belong to the same event type, the event data may reflect different urban operation problems. For example, events related to holiday management can be divided into issues such as the opening time of schools in high temperatures and holiday adjustment issues. If these issues are mixed together, it may be difficult to mine hotspot information or the mined hotspot information may be biased due to the different focuses of the issues.

[0073] Therefore, in order to more accurately mine hotspot information based on event type, this embodiment can perform first-level clustering of multiple event data according to event type and semantic similarity. Here, clustering based on event semantics can more accurately group events describing different urban operational problems into different clusters, thereby more effectively mining potential hotspot information.

[0074] For example, suppose multiple event data include event data belonging to three event types: A, B, and C. Event data of type A are denoted as A1, A2, ..., and so on. Similarly, event data of type B are denoted as B1, B2, etc., and event data of type C are denoted as C1, C2, etc. Then, after the first level of clustering, the resulting event clusters can include: cluster 1 [A1, A2, ..., A30], cluster 2 [A31, A32, ..., A37], cluster 3 [B1, B2, ..., B10], cluster 4 [B11, B12, ..., B25], cluster 5 [B26, B27], and cluster 6 [C1, C2, ..., C5].

[0075] It is understandable that clustering based on event type and semantic similarity may result in some event clusters containing very little event data. A necessary condition for an event to become a hot event is that the number of events describing the same problem is greater than a certain threshold. Therefore, in order to accurately identify hot events in the event type dimension, this embodiment can determine whether the number of events in each event cluster in at least one event cluster is greater than a preset first threshold. If so, the event cluster is determined to be a hot event cluster; otherwise, the event cluster is determined not to be a hot event cluster.

[0076] After passing through the first threshold screening, a large number of events that are of concern to citizens can be effectively identified, namely hot topics.

[0077] It should be noted that the first threshold mentioned above can be determined according to the actual scenario. For example, if there are many events in the event list, a larger first threshold can be set, and vice versa. Of course, there are other ways to determine the first threshold, which are not specifically limited here.

[0078] For example, if the first threshold is 5, the following hot event clusters are obtained after filtering: cluster 1 [A1, A2, ..., A30], cluster 2 [A31, A32, ..., A37], cluster 3 [B1, B2, ..., B10], cluster 4 [B11, B12, ..., B25], and cluster 6 [C1, C2, ..., C5].

[0079] Step S203: Perform second-level clustering of hot event clusters according to the involved subjects to obtain at least one event sub-cluster, and take the event sub-cluster with the number of events in at least one event sub-cluster greater than the second threshold as the hot event sub-cluster.

[0080] Considering that a cluster of hot events may be associated with multiple involved entities, in order to clarify the hot events at the entity level, this embodiment can also perform a second-level clustering of each hot event cluster according to the involved entities, obtaining at least one event sub-cluster. Here, each event sub-cluster corresponds to one involved entity.

[0081] For example, after clustering the aforementioned hot event clusters by the involved entities, the following event subclusters may be obtained: Subcluster 1 [A1, A2, ..., A6], Subcluster 2 [A7, A8, ..., A30], Subcluster 3 [A31, A32, A33], Subcluster 4 [A34, A35, A36, A37], Subcluster 5 [B1, B2, ..., B10], Subcluster 6 [B11, B12, ..., B25], Subcluster 7 [C1, C2, ..., C4], Subcluster 8 [C5].

[0082] Similarly, in order to identify hot events in the dimension of the involved subject, this embodiment can determine whether the number of events in each event sub-cluster in at least one event sub-cluster is greater than a preset second threshold. If so, the event sub-cluster is determined to be a hot event sub-cluster; otherwise, the event sub-cluster is determined not to be a hot event sub-cluster.

[0083] For example, if the second threshold is 5, then the hotspot event subclusters finally determined in this embodiment include: subcluster 1 [A1, A2, ..., A6], subcluster 2 [A7, A8, ..., A30], subcluster 5 [B1, B2, ..., B10], and subcluster 6 [B11, B12, ..., B25].

[0084] Step S204: Generate hotspot information based on hotspot event clusters and hotspot event subclusters.

[0085] Here, "hot topic information" refers to information describing trending events.

[0086] In this embodiment, each hot event cluster points to a hot event in the event type dimension, and each hot event sub-cluster points to a hot event in the subject dimension. Based on this, the hot event information includes: information describing hot events in the event type dimension, and information describing hot events in the subject dimension.

[0087] The hot topic information mining method provided in this application, in order to discover hot events within a required time period, labels multiple event data within the required time period to obtain event tags corresponding to each event data. These event tags include event type and involved parties. Next, a first-level clustering is performed on the multiple event data based on event type and semantic similarity, resulting in at least one event cluster. Event clusters with more than a first threshold number of events are designated as hot topic event clusters. Each event cluster within the at least one event cluster corresponds to the same event type. Clustering by event type groups events of the same type together, avoiding noise interference from different types of events in the hot topic information mining process. Furthermore, considering that even events belonging to the same event type may reflect different urban operation problems, and a hot topic event only focuses on one type of urban operation problem, semantic similarity clustering is also introduced during event type clustering. This more accurately clusters events reflecting urban operation problems repeatedly mentioned by citizens, improving the accuracy of hot topic information mining and reducing the risk of missing or biased hot topic information.

[0088] Furthermore, considering that a cluster of hot events may be associated with multiple stakeholders, in order to clarify the hot events at the stakeholder level, this application can further perform a second-level clustering of the hot event clusters by stakeholder, obtaining at least one event sub-cluster, and designating the event sub-cluster with the number of events exceeding a second threshold as the hot event sub-cluster. Finally, hot event information is generated based on the hot event clusters and hot event sub-clusters. Therefore, this application considers not only hot events at the event type level but also hot events at the stakeholder level, making the final generated hot event information more comprehensive. By identifying hot events at the stakeholder level, this application can more accurately determine the objects that pose potential risks to social operations, thereby enabling more targeted responses, reducing risk management costs, and improving urban operational efficiency.

[0089] The following detailed description of each step is provided through a specific embodiment. It should be noted that the following embodiment is merely an example and is not intended to limit the scope of this application.

[0090] In one possible implementation, in order to tag multiple event data within the required time period in step S201, this embodiment can pre-configure a classification reasoning engine and a first large language model. The classification reasoning engine is used to classify the multiple event data into event types, and the first large language model is used to extract target event elements from the multiple event data, which include the parties involved. Thus, the multiple event data can be tagged according to the event types and target event elements.

[0091] Optionally, the aforementioned classification inference engine classifies event data based on an event classification model. To enable the classification inference engine to classify event data more accurately and obtain its corresponding event type, this embodiment can pre-train the model file used by the classification inference engine. Specifically, this embodiment can train the event classification model based on training event data and its corresponding event type labels to obtain the aforementioned model file. Here, the event classification model refers to the model pre-built for the classification inference engine.

[0092] To obtain highly accurate event type labels, one possible approach is manual labeling, whereby individuals manually identify the real event types corresponding to the training event data from the event type field.

[0093] Model training requires a large amount of training data. Labeling large-scale training event data requires a lot of manual time and effort, which is inefficient and may lead to labeling errors because some event types are semantically similar.

[0094] To improve the efficiency and accuracy of event type labeling, this embodiment also provides the following optional methods.

[0095] First, the event classification model pre-built for the classification inference engine can be used to classify and predict the training event data, thereby obtaining the event type prediction results corresponding to the training event data.

[0096] It is understandable that when the event classification model in the classification inference engine is not trained or is undertrained, it may obtain incorrect event type prediction results. In order to avoid training the event classification model with incorrect event type labels, this embodiment can leverage the powerful semantic understanding and reasoning capabilities of the large language model, use the configured second large language model to determine whether the training event data belongs to the event type prediction result corresponding to the training event data, and generate suggested event types when the determination result is negative.

[0097] For example, by configuring an event classification prompt instruction template for a second language model through prompting engineering, this template includes event type dictionary filling slots, event data filling slots, and event type filling slots. The template instructs the second language model to determine whether the training event data in the event data filling slot belongs to the event type prediction result in the event type filling slot. If not, it selects and outputs a suggested event type consistent with the training event data from the event type dictionary filling slot. Thus, by filling the event data filling slot with training event data, the event type dictionary filling slot with the event type dictionary filling slot, and the event type prediction result corresponding to the training event data filling slot with the event type filling slot, the event classification prompt instruction can be obtained. Inputting this instruction into the second language model yields the judgment result and / or suggested event type output by the second language model.

[0098] Furthermore, the judgment results and / or suggested event types from the second largest language model are output and displayed so that humans can analyze them based on the displayed judgment results and / or suggested event types and return the actual event types. If the judgment result is yes, the actual event type is the event type prediction result output by the classification reasoning engine. If the judgment result is no, the actual event type is the event type suggested by the second largest language model, or other event types provided by humans after analyzing the suggested event types.

[0099] Therefore, in this embodiment, real event types and training event data can be used as training data, that is, training event data can be used as data samples and real event types can be used as sample labels to train the event classification model and obtain the trained model file.

[0100] Optionally, this embodiment may extract 100-200 event data points for each event type as training data. Of course, other amounts of training data may also be used, and this application does not impose specific limitations.

[0101] Optionally, in order for the training data to be recognized by the classification inference engine, this embodiment can first convert the training data into the input parameter format file required by the engine after obtaining the training data, and then input it into the classification inference engine (the engine at this time can be understood as a classification training engine) to train the event classification model.

[0102] Optionally, when inputting machine-readable training data into the classification inference engine, a callback address after training is completed can also be input so that the caller can be notified of the current training status in a timely manner through the callback address.

[0103] When the training status is "training successful", the trained model file can be replaced in the directory of the classification inference engine and the classification inference engine can be restarted so that the classification inference engine can classify and label multiple event data within the required time period.

[0104] Optionally, the steps to restart the classification inference engine are as follows: Use the command `lsof -i:8200` to find the process occupying port 8200, obtain the process ID of the classification inference engine service, and use `kill -9 PID` to forcibly stop the process. Then, change the environment variables, such as `conda activate znfp`, switch to the startup script directory of the classification inference engine service, and run the test script `client_test.py`. If the interface returns a 200 status code, it means that the service has been restarted successfully.

[0105] Compared to manually selecting event types directly from an event type dictionary, this embodiment, with the assistance of a classification reasoning engine and a second language model, can help humans provide more accurate feedback on real event types, thereby improving the classification accuracy of the classification reasoning engine.

[0106] Optionally, the process of the first language model extracting target event elements from multiple event data can be achieved by pre-writing an element extraction prompt instruction template. This template includes an event data filling slot, in which one or more event data can be filled to form an element extraction prompt instruction. This instruction can then be input into the first language model to obtain the target event elements output by the model.

[0107] In this embodiment, the target event element includes: the involved entity, which may optionally include the full name and / or abbreviation of the involved entity.

[0108] For example, if the full name of the entity involved is a four-part structure including the administrative region (such as province), the trade name (a unique and identifiable name used by the enterprise to distinguish itself from other peers in its business activities), the business characteristics (such as construction labor), and the organizational form (such as limited liability company), then the entity involved can be simply referred to as a two-part structure including the trade name and business characteristics.

[0109] Of course, the abbreviation of the entity involved can also be in other forms, depending on the specific circumstances.

[0110] For example, the prompt text in the feature extraction prompt instruction template can be as follows:

[0111] "You are a government citizen service hotline expert. You possess relevant knowledge in the field of citizen hotline complaints and are able to apply this knowledge to analyze hotline complaints. Based on your professional knowledge and experience, please complete the following two tasks:"

[0112] 1. Identify “mainBody (full name of the entity involved)”: Extract the information of the “complaint target” that ultimately points to the event data.

[0113] 2. Extract “simpleBody (abbreviation of the subject of the complaint)”: Based on “mainBody”, extract keywords of “the subject of the complaint”.

[0114] Require:

[0115] 1. "mainBody" and "simpleBody" should only output the responsible parties related to the citizens' demands, and should not output other subject information.

[0116] 2. Please use JSON format for output.

[0117] 3. There are multiple sets of "mainBody" and "simpleBody" in the event data. Please output them one by one.

[0118] It should be noted that, in addition to the parties involved, the target event elements may also include other information, such as a brief description of the event data.

[0119] It should also be noted that the first and second major language models mentioned above can be the same major language model or different major language models. Specifically, they can be any major language model such as the Wenxin Yiyan major model, the Xunfei Xinghuo major model, the Deep Search major model, or a major language model that will emerge in the future. This application does not impose any specific limitations.

[0120] In one possible implementation, the event tag corresponding to the above event data may also include the latitude and longitude coordinates of the location where the event data occurred. Then, the latitude and longitude coordinates of the location where the event data occurred can be obtained by querying the latitude and longitude coordinate alignment interface.

[0121] For example, see Figure 3 The diagram shown is a schematic representation of an event data tagging process provided in this application. Figure 3 The top part is a description of the event with the event number "240131T09443". The bottom part is the event tags corresponding to the event data of the event "240131T09443", including: event type (skills training service dispute), involved parties (abcd), and latitude and longitude coordinates (longitude 104.16022200, latitude 30.702740).

[0122] Of course, event tags can be other than those specified in this application.

[0123] After the events are tagged, the first-level clustering can be performed through S202 to obtain at least one event cluster.

[0124] Optionally, to ensure that each event cluster corresponds to the same event type, this embodiment can first cluster multiple event data according to event type to obtain at least one first cluster. Here, the event data in the first cluster correspond to the same event type. Then, the first cluster with more than a third threshold in the at least one first cluster is taken as the second cluster. The event semantic similarity of the event data in the second cluster is calculated pairwise, and the event data in the second cluster is clustered according to the calculated event semantic similarity to obtain at least one third cluster, which is taken as at least one event cluster.

[0125] This embodiment uses a third threshold for initial screening, which avoids subsequent semantic classification of the first cluster with a small number of events, thus improving the efficiency of event clustering. The third threshold can be determined based on the actual scenario; for example, in some scenarios, the third threshold can be 3.

[0126] Optionally, if the target event elements mentioned above include a brief description of the event data, the process of "calculating the semantic similarity of each pair of event data in the second cluster" may include: using the configured deduplication engine, for each pair of event data in the second cluster, determining the event time, event location, and reporter identifier of each of the two event data, and calculating the similarity of the brief description, event time, event location, and reporter identifier of each of the two event data, and determining the semantic similarity of the two event data based on the calculated similarity.

[0127] Preferably, if the location of the incident includes the aforementioned latitude and longitude coordinates, the similarity of the locations is calculated based on the latitude and longitude coordinates; if the location of the incident does not include latitude and longitude coordinates, the text of the location (e.g., No. 3, xx Road) is converted into a vector representation, and then the vector distance is calculated to obtain the similarity of the locations.

[0128] Optionally, this embodiment can pre-configure weights for the content summary, the time of the incident, the location of the incident, and the reporter's identifier (such as the reporter's name), and then perform a weighted sum based on the configured weights and the calculated similarity to obtain the event semantic similarity between the two event data.

[0129] The weights here can be determined based on the actual scenario. For example, in one possible scenario, considering that the content summary describes the overall outline of the event, its similarity can better reflect the semantic relationship between the two event data. Therefore, a larger weight, such as 0.9, can be assigned to the content summary, and the remaining similarities can be evenly assigned a total weight of 0.1.

[0130] Optionally, the process of "clustering event data within the second cluster based on the calculated event semantic similarity to obtain at least one third cluster" can be implemented in various ways, and this application provides, but is not limited to, the following:

[0131] The first method involves selecting one event data point from the second cluster, clustering all event data points in the second cluster with an event semantic similarity greater than or equal to a preset similarity threshold into a third cluster, and removing the event data points from the third cluster from the second cluster. Then, another event data point is selected from the remaining event data points in the second cluster, clustering all event data points in the second cluster with an event semantic similarity greater than or equal to a preset similarity threshold into a new third cluster, and removing the event data points from the new third cluster from the second cluster. This process continues until there is no event data in the second cluster.

[0132] The second approach involves assembling the semantic similarity of events between pairs of events in the second cluster into a similarity matrix, and then using any one of hierarchical clustering, K-means clustering, or DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to obtain at least one third cluster.

[0133] Of course, there are other ways to implement this, which will not be listed here.

[0134] This embodiment calculates the semantic similarity of event data in multiple dimensions, including content summary, time of occurrence, location of occurrence, and reporter identifier, making the calculated semantic similarity more accurate and thus improving the accuracy of the first-level clustering.

[0135] Further, perform the second-level clustering according to step S203.

[0136] Considering that citizens (reporters) and staff compiling the event list may not be aware of the full names of the entities involved, resulting in incomplete descriptions of the entities in the reported event data, and that different citizens may provide different descriptions of the same entity, for example, for "xx (province) xx (trade name) Cultural Media Co., Ltd.", citizen A reports the entity as "xx (trade name) Cultural Media Co., Ltd.", citizen B reports it as "xx (province) xx (trade name) Cultural Media Company", and citizen C reports it as "xx (trade name) Cultural Media". To cluster event data reflecting the same entity into a single cluster, optionally, when clustering by entity in step S203, clustering can be performed based on the brief descriptions of the entities extracted earlier.

[0137] Compared to clustering by the full name of the involved entity or by the directly reported involved entity, clustering by the abbreviation of the involved entity can more accurately group related events of the same involved entity together, thus improving the accuracy of subsequent hot topic information mining.

[0138] The following describes the process of step S204, "generating hotspot information based on hotspot event clusters and hotspot event subclusters".

[0139] Considering that citizens may provide incorrect event locations when reporting event data, such as reporting an event that occurred at store one as occurring at store two, the hotspot event subclusters may contain event data with incorrect clustering due to location errors. To avoid these noisy event data with clustering errors adversely affecting the generated hotspot information, this embodiment can optionally determine the event location of each event data in the hotspot event subcluster before generating hotspot information. Then, a baseline event data is selected from the hotspot event subcluster and placed in the fourth cluster. The other event data in the hotspot event subcluster, excluding the baseline event data, are traversed: the distance between the baseline event data and the event location of each of the currently traversed event data is determined. If the distance is less than or equal to a preset distance threshold, the currently traversed event data is placed in the fourth cluster. At the end of the traversal, it is determined whether the number of events in the fourth cluster is greater than the target number, which is determined based on the number of events in the hotspot event subcluster. If not, the fourth cluster is cleared, and a baseline event data is selected again from the hotspot event subcluster and placed in the fourth cluster. The traversal of other event data in the hotspot event subcluster is then resumed. If yes, hotspot information is generated based on the hotspot event subcluster and the fourth cluster.

[0140] For example, if a hotspot event sub-cluster includes event data 1-10, event data 1 can be used as the baseline event data and placed in the fourth cluster. Then, the distance (e.g., Euclidean distance) between the locations of events 2-10 and the location of event data 1 can be calculated. Event data whose distance from the location of event data 1 is less than a preset distance threshold (e.g., 2 kilometers) can also be placed in the fourth cluster.

[0141] Considering that location reporting errors are a low-probability event, if the number of events in the fourth cluster is greater than the target number (e.g., half or 80% of the number of events in the hotspot event sub-cluster), it indicates that there are no location reporting errors in the event data within the fourth cluster. In this case, hotspot information is generated based on the fourth cluster and the hotspot event cluster. Conversely, if the number of events in the fourth cluster is less than the target number (e.g., half or 80% of the number of events in the hotspot event sub-cluster), it indicates that the base event data itself may have location reporting errors. In this case, the fourth cluster can be cleared, and event data 2 can be used as the base event data, repeating the above process.

[0142] By using the above method, event data with location reporting errors can be removed from the hotspot event sub-clusters, thereby improving the accuracy of subsequent hotspot information.

[0143] In one possible implementation, this embodiment can utilize a configured third language model to generate hot topic titles and descriptions based on hot topic event clusters and / or a fourth cluster, and determine the summary information on hot topic event handling corresponding to the hot topic event clusters and / or the fourth cluster. Here, the third language model can be the same as or different from the first or second language model mentioned above; this application does not impose any limitations.

[0144] The above-mentioned summary information on the handling of hot issues refers to the relevant information on the handling results of the hot issues identified in the analysis by the relevant departments.

[0145] This embodiment can also determine the influence measurement values ​​corresponding to the hot event cluster and the fourth cluster respectively. Optionally, the influence measurement values ​​include one or more of the following values: duration, number of events, number of complainants, proportion of hot events, and number of related parties involved.

[0146] Therefore, the hot topic titles, descriptions, summary information on handling hot topics, and influence metrics corresponding to the hot topic event clusters and the fourth cluster are used as hot topic information.

[0147] See Figure 4 This is a diagram illustrating the hotspot information corresponding to a cluster of hot events. For example... Figure 4 This embodiment can generate a hot topic title for each hot topic event cluster, and determine the number of relevant involved entities, the number of complainants, the number of events, the proportion of hot topic events, and the analysis period. In addition, it also marks the event type corresponding to the hot topic event cluster.

[0148] See Figure 5 This diagram illustrates the hot topic information corresponding to the fourth cluster of the "school opening time issue during high temperatures" topic. For example... Figure 5 The hot topic information generated for each involved party includes: number of complainants, number of incidents, time percentage, and analysis period.

[0149] certainly, Figure 4 and Figure 5 The hotspot information shown is for illustrative purposes only and is not intended to limit this application.

[0150] In one possible implementation, an analysis report can also be generated based on hotspot information, such as... Figure 6 The document presents an analysis report that may include: hotspot description information, extract of involved entities (preferably the full names of the involved entities), event type, relevant event statistics, event location, event handling analysis (i.e., summary information on the handling of hotspot events), and hotspot duration analysis.

[0151] As introduced above, in this embodiment, the hot event clusters correspond to hot events at the event type level, and the fourth cluster corresponds to hot events at the subject level. Analyzing hot events can promptly identify potential urban operational risks. For example, the hot event "schools delaying opening due to high temperatures" could lead to heatstroke among teachers and students. To minimize this risk, this embodiment can retrieve knowledge fragments related to the hot event information from a pre-configured knowledge base, and then generate corresponding handling measures for the hot event clusters and / or the fourth cluster based on the knowledge fragments and the hot event information. The aforementioned knowledge base is built based on data related to people's livelihoods, such as national or local policies and social governance data.

[0152] For example, see Figure 7 This is a schematic diagram illustrating proposed measures to address the issue of delayed school opening at Jianshe Road Primary School due to high temperatures. The diagram shows measures from five perspectives: assessing weather conditions, developing emergency plans, strengthening public education, optimizing the campus environment, and establishing a feedback mechanism. These measures can more effectively prevent heatstroke among teachers and students.

[0153] Of course, the disposal measures can also be included as part of the analysis report, and this application does not impose specific limitations.

[0154] In summary, this embodiment, through technologies such as classification reasoning engine, knowledge enhancement, and large model prompt word engineering, can more accurately identify the development trend of events in specific times and regions, and comprehensively analyze hot events. This enables the discovery and handling of urban risk warnings to shift from passive to proactive, and from manual discovery to intelligent perception, thereby reducing risk management costs and improving urban operational efficiency.

[0155] The above describes a hotspot information mining method provided by the embodiments of this application. The following will describe the apparatus for performing the above hotspot information mining method.

[0156] Please see Figure 8 , Figure 8 This is a schematic diagram of a hotspot information mining device provided in an embodiment of this application. Figure 8 As shown, the hotspot information mining device may include:

[0157] The event tagging unit 801 is used to tag multiple event data within the required time period to obtain event tags corresponding to the multiple event data. The event tags include the event type and the involved parties.

[0158] The first clustering unit 802 is used to perform first-level clustering on multiple event data according to event type and event semantic similarity to obtain at least one event cluster. Event clusters with more than one threshold of events in at least one event cluster are designated as hot event clusters. Event data in each event cluster correspond to the same event type.

[0159] The second clustering unit 803 is used to perform second-level clustering of hot event clusters according to the subjects involved, to obtain at least one event sub-cluster, and to take the event sub-cluster with the number of events in the at least one event sub-cluster that is greater than a second threshold as the hot event sub-cluster.

[0160] Hotspot information generation unit 804 is used to generate hotspot information based on hotspot event clusters and hotspot event subclusters.

[0161] Each module in the aforementioned hotspot information mining device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0162] This application also provides an electronic device, which may include at least one processor and a memory connected to the processor, wherein:

[0163] Memory is used to store computer programs;

[0164] The processor is used to execute computer programs to enable electronic devices to implement any of the hotspot information mining methods provided in the embodiments of this application.

[0165] refer to Figure 9The diagram illustrates a structural schematic suitable for implementing the electronic device in the embodiments of this application. The electronic device in the embodiments of this application may include, but is not limited to, fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 9 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0166] like Figure 9 As shown, the electronic device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage device 908 into a random access memory (RAM) 903. When the electronic device is powered on, the RAM 903 also stores various programs and data required for the operation of the electronic device. The processing unit 901, ROM 902, and RAM 903 are interconnected via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0167] Typically, the following devices can be connected to I / O interface 905: input devices 906 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 907 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 908 including, for example, memory cards, hard drives, etc.; and communication devices 909. Communication device 909 allows electronic devices to exchange data via wireless or wired communication with other devices. Although Figure 9 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.

[0168] This application also provides a computer program product including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the hotspot information mining methods provided in this application.

[0169] This application also provides a computer-readable storage medium that carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the hotspot information mining methods provided in this application.

[0170] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0171] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0172] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0173] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A method for mining hotspot information, characterized in that, include: Multiple event data within the required time period are tagged to obtain event tags corresponding to the multiple event data, and the event tags include event type and involved parties; Based on the event type and event semantic similarity, the multiple event data are clustered at the first level to obtain at least one event cluster. The event clusters with more than one threshold of events in the at least one event cluster are regarded as hot event clusters. The event data in each event cluster correspond to the same event type. The hot event clusters are clustered at the second level according to the involved entities to obtain at least one event sub-cluster. The event sub-clusters in which the number of events is greater than a second threshold are regarded as hot event sub-clusters. Hotspot information is generated based on the hotspot event clusters and the hotspot event subclusters.

2. The hotspot information mining method according to claim 1, characterized in that, The tagging of multiple event data within the required time period includes: The configured classification reasoning engine is used to classify the multiple event data into event types, thereby obtaining the event types corresponding to the multiple event data. Using the configured first language model, target event elements are extracted from the multiple event data respectively to obtain the target event elements corresponding to the multiple event data, and the target event elements include the involved subject; The event data are labeled according to the event type and target event element corresponding to each of the event data.

3. The hotspot information mining method according to claim 2, characterized in that, The training process for the model files used by the classification inference engine includes: Using the event classification model pre-built for the classification inference engine, the training event data is classified and predicted to obtain the event type prediction result corresponding to the training event data; The second language model is configured to determine whether the training event data belongs to the event type prediction result corresponding to the training event data, and if the determination result is negative, a suggested event type is generated. The judgment results and / or the suggested event types from the second largest language model are output and displayed to obtain the real event types from human feedback; The event classification model is trained using the real event types and the training event data to obtain the trained model file.

4. The hotspot information mining method according to claim 2, characterized in that, The first-level clustering of the multiple event data based on the event type and event semantic similarity to obtain at least one event cluster includes: The multiple event data are clustered according to the event type to obtain at least one first cluster, and the event data in the first cluster correspond to the same event type; The first cluster in which the number of events in at least one first cluster is greater than the third threshold is designated as the second cluster; Calculate the semantic similarity of each pair of event data within the second cluster, and cluster the event data within the second cluster based on the calculated semantic similarity to obtain at least one third cluster, which serves as the at least one event cluster.

5. The hotspot information mining method according to claim 4, characterized in that, The target event element also includes a brief description of the event data. The step of calculating the semantic similarity of each pair of event data within the second cluster includes: Using the configured deduplication engine, for every two event data in the second cluster, the event time, event location, and reporter identifier of each of the two event data are determined, and the similarity of the content summary, event time, event location, and reporter identifier of each of the two event data is calculated. Based on the calculated similarity, the event semantic similarity of the two event data is determined.

6. The hotspot information mining method according to claim 1, characterized in that, The step of generating hotspot information based on the hotspot event clusters and the hotspot event sub-clusters includes: Determine the location of each event data point within the hotspot event sub-cluster; Select a baseline event data point from the hotspot event sub-cluster and place it in the fourth cluster. Then, iterate through the other event data within the hotspot event sub-cluster. Determine the distance between the location of the baseline event data and the location of the currently traversed event data. If the distance is less than or equal to a preset distance threshold, then place the currently traversed event data in the fourth cluster. At the end of the traversal, it is determined whether the number of events in the fourth cluster is greater than the target number, which is determined based on the number of events in the hot event sub-cluster. If not, clear the fourth cluster, and select a new base event data from the hot event sub-cluster and place it in the fourth cluster, then return the other event data within the hot event sub-cluster. If so, then hotspot information is generated based on the hotspot event cluster and the fourth cluster.

7. The hotspot information mining method according to claim 6, characterized in that, The step of generating hotspot information based on the hotspot event cluster and the fourth cluster includes: Using the configured third language model, generate hot topic titles and descriptions based on the hot topic event clusters and / or the fourth cluster, and determine the summary information of hot topic event handling corresponding to the hot topic event clusters and / or the fourth cluster; Determine the influence metric values ​​corresponding to the hot topic event cluster and the fourth cluster, respectively; The hot topic information is defined as the hot topic title, hot topic description information, hot topic event handling summary information, and influence measurement value corresponding to the hot topic event cluster and the fourth cluster, respectively.

8. The hotspot information mining method according to claim 7, characterized in that, Also includes: Retrieve knowledge fragments related to the hot topic information from a pre-configured knowledge base; Based on the knowledge fragments and the hotspot information, generate the hotspot event clusters and / or the corresponding processing measures for the fourth cluster.

9. The hotspot information mining method according to any one of claims 1-8, characterized in that, The entities involved are referred to as "the entities involved" in this context.

10. A computer program product, characterized in that, It includes computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the hotspot information mining method as described in any one of claims 1 to 9.

11. An electronic device, characterized in that, It includes at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program so that the electronic device can implement the hotspot information mining method as described in any one of claims 1 to 9.

12. A computer storage medium, characterized in that, The storage medium carries one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the hotspot information mining method as described in any one of claims 1 to 9.

Images