Artificial intelligence-driven public place risk dynamic early warning method and device, equipment and medium

Abstract

The embodiment of the invention provides an artificial intelligence-driven public place risk dynamic early warning method and device, equipment and a medium. The method comprises the steps of collecting text release information associated with a target place, and performing text clustering on the text release information to obtain a target clustering result; analyzing sub-class clusters in the target clustering result to obtain related feature information; performing topic identification on the target clustering result according to the related feature information to obtain an initial topic and related text information; performing association analysis according to the related text information and the target place to obtain a target topic and associated text information; performing behavior detection on the associated text information to obtain a target behavior type; when the target behavior type meets a preset type, collecting latest text information associated with the target place; performing popularity analysis on the target topic according to the latest text information and the related text information to obtain a target popularity type; and performing risk early warning on the target place according to the target popularity type and the target behavior type to obtain a target early warning result.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-driven public place risk dynamic early warning method, device, equipment and medium. Background Art

[0002] With the development of the Internet, more and more users are keen to share their behavior in public places on the Internet. Whether it is extreme sports challenges in a busy shopping mall or trying new and dangerous games on a busy street, these contents have become eye-catching materials on the Internet. However, when some users make dangerous behaviors in public places and make them public, some people who lack judgment and safety awareness will be influenced by these contents and imitate these dangerous behaviors. This imitation behavior grows exponentially under the communication effect of the Internet, which greatly increases the safety risks in public places. However, under the existing technical system, most public places’ safety risk warnings mainly rely on the information captured by the cameras installed in the venues. They can capture some obvious dangerous behaviors, such as fights, fire smoke and other abnormal conditions, and issue alarms in time, which plays an important role in ensuring the safety of public places. However, the existing technology lacks the ability to timely discover potential safety risks from public information on the Internet, which greatly reduces the timeliness of risk warnings. Summary of the invention

[0003] The main purpose of the embodiments of the present invention is to provide an artificial intelligence-driven dynamic early warning method, device, equipment and medium for risks in public places, aiming to solve the problem that related technologies lack the ability to promptly discover potential security risks from public information on the Internet, thereby reducing the timeliness of risk warnings.

[0004] In a first aspect, an embodiment of the present invention provides an artificial intelligence-driven public place risk dynamic early warning method, comprising:

[0005] Collecting text release information associated with the target place at the first time, and performing text clustering on the text release information to obtain a target clustering result;

[0006] Performing feature analysis on each subclass cluster in the target clustering result to obtain corresponding relevant feature information;

[0007] Performing topic identification on the target clustering result according to the relevant feature information to obtain an initial topic and relevant text information corresponding to the initial topic;

[0008] Performing association analysis based on the relevant text information and the target place to obtain a target topic corresponding to the target place and associated text information corresponding to the target topic;

[0009] Conduct behavior detection on the associated text information to obtain the target behavior type corresponding to the target topic;

[0010] When the target behavior type meets the preset type, collect the latest text information associated with the target venue at the second time;

[0011] Conduct heat analysis on the target topic based on the latest text information and the relevant text information to obtain the target heat type corresponding to the target topic;

[0012] Conduct dynamic risk warning on the target venue based on the target heat type and the target behavior type to obtain the target warning result corresponding to the target venue.

[0013] In a second aspect, an embodiment of the present invention provides an artificial intelligence-driven dynamic risk warning device for public places, including:

[0014] A data collection module, configured to collect text publication information associated with a target venue at a first time, and perform text clustering on the text publication information to obtain a target clustering result;

[0015] A feature analysis module, configured to perform feature analysis on each sub-cluster in the target clustering result to obtain corresponding relevant feature information;

[0016] A topic recognition module, configured to perform topic recognition on the target clustering result based on the relevant feature information to obtain an initial topic and the relevant text information corresponding to the initial topic;

[0017] An association analysis module, configured to perform association analysis on the relevant text information and the target venue to obtain the target topic corresponding to the target venue and the associated text information corresponding to the target topic;

[0018] A behavior detection module, configured to perform behavior detection on the associated text information to obtain the target behavior type corresponding to the target topic;

[0019] A data judgment module, configured to collect the latest text information associated with the target venue at a second time when the target behavior type meets the preset type;

[0020] A heat analysis module, configured to perform heat analysis on the target topic based on the latest text information and the relevant text information to obtain the target heat type corresponding to the target topic;

[0021] A risk warning module, configured to perform dynamic risk warning on the target venue based on the target heat type and the target behavior type to obtain the target warning result corresponding to the target venue.

[0022] In a third aspect, an embodiment of the present invention further provides a terminal device, which includes a processor, a memory, a computer program stored on the memory and executable by the processor, and a data bus for realizing connection communication between the processor and the memory. When the computer program is executed by the processor, the steps of any one of the artificial intelligence-driven public place risk dynamic early warning methods provided in the specification of the present invention are implemented.

[0023] In a fourth aspect, an embodiment of the present invention further provides a storage medium for computer-readable storage. The storage medium is characterized in that it stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any one of the artificial intelligence-driven public place risk dynamic early warning methods provided in the specification of the present invention.

[0024] An embodiment of the present invention provides an artificial intelligence-driven public place risk dynamic early warning method, device, equipment and medium. The method includes: collecting text publishing information associated with a target place at a first time, performing text clustering on the text publishing information to obtain a target clustering result, and performing feature analysis on each sub-cluster in the target clustering result to obtain corresponding relevant feature information; performing topic recognition on the target clustering result according to the relevant feature information to obtain an initial topic and relevant text information corresponding to the initial topic; performing association analysis on the relevant text information and the target place to obtain a target topic corresponding to the target place and associated text information corresponding to the target topic. This association analysis can deeply explore the internal connection between the text information and the target place, providing a more accurate basis for subsequent risk assessment; then performing behavior detection on the associated text information to obtain a target behavior type corresponding to the target topic; when the target behavior type meets a preset type, collecting the latest text information associated with the target place at a second time; performing heat analysis on the target topic according to the latest text information and the relevant text information to obtain a target heat type corresponding to the target topic, and then performing dynamic risk early warning on the target place according to the target heat type and the target behavior type, so as to be able to evaluate the safety risk status of the target place in real time and accurately, and then obtain a target early warning result. This method can timely discover the target behavior type in the target place from the text publishing information published by users, and then timely perform risk early warning according to the target behavior type and the target heat type, thus significantly improving the safety of public places. This method makes full use of the value of network public information, making up for the shortcoming in related technologies of lacking the ability to timely discover potential safety risks from network public information. In the past, due to the inability to timely obtain various feedback information about public places on the network, many safety problems were often discovered when they developed to a more serious level. And this method can insight potential safety risks in advance, realizing the transformation from passive response to active prevention, providing a strong guarantee for the safe operation of public places, and making people feel more at ease when moving in public places. Brief Description of the Drawings

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0026] Figure 1 It is a schematic flowchart of a method for dynamically warning risks in public places driven by artificial intelligence provided by an embodiment of the present invention;

[0027] Figure 2 It is a schematic block diagram of the module structure of a device for dynamically warning risks in public places driven by artificial intelligence provided by an embodiment of the present invention;

[0028] Figure 3 It is a schematic block diagram of the structure of a terminal device provided by an embodiment of the present invention. Detailed Embodiments

[0029] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the protection scope of the present invention.

[0030] The flowchart shown in the drawings is only an example, and does not necessarily include all contents and operations / steps, nor does it necessarily execute in the described order. For example, some operations / steps can be decomposed, combined, or partially merged, so the actual execution order may change according to the actual situation.

[0031] It should be understood that the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0032] The embodiments of the present invention provide a method, device, equipment and medium for dynamically warning risks in public places driven by artificial intelligence. Among them, the method for dynamically warning risks in public places driven by artificial intelligence can be applied to a terminal device, which can be an electronic device such as a tablet computer, a notebook computer, a desktop computer, a personal digital assistant, and a wearable device. The terminal device can be a server or a server cluster.

[0033] The following will describe in detail some embodiments of the present invention with reference to the accompanying drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0034] Please refer to Figure 1 , Figure 1 , which is a schematic flowchart of an artificial intelligence-driven dynamic risk early warning method for public places provided by an embodiment of the present invention.

[0035] As Figure 1 shown, the artificial intelligence-driven dynamic risk early warning method for public places includes steps S101 to S108.

[0036] Step S101: Collect text release information associated with the target place at the first time, and perform text clustering on the text release information to obtain a target clustering result.

[0037] Exemplarily, the target place is any one of the public places where risk early warning is required, and the target place can be a park or a popular scenic spot, etc. The first time is a historical time, and then relevant text release information publicly released by relevant users associated with the target place at the first time is crawled according to web crawler technology. The text release information is information related to the target place published by relevant users or information corresponding to the information position when the information published by relevant users is located at the target place.

[0038] Exemplarily, text clustering is performed on the text release information according to a text clustering algorithm such as the K-means clustering algorithm to obtain a target clustering result.

[0039] In some embodiments, the performing text clustering on the text release information to obtain a target clustering result includes: preprocessing the text release information to obtain initial keywords corresponding to the text release information; obtaining the word frequency corresponding to the initial keywords from the text release information, and determining the first text quantity corresponding to the inclusion of the initial keywords from all the text release information; classifying the text type of the text release information to obtain the text type corresponding to the text release information; associating the initial keywords with the text type to obtain the second text quantity corresponding to the inclusion of the initial keywords under each text type; obtaining the maximum value among all the second text quantities to obtain the maximum text quantity, and then determining the word importance corresponding to the initial keywords according to the word frequency in combination with the first text quantity and the maximum text quantity; screening the initial keywords according to the word importance to obtain target keywords corresponding to the text release information; performing text clustering on the text release information according to the target keywords to obtain the target clustering result; wherein, the word importance is obtained according to the following formula:

[0040]

[0041] in, Indicates the word importance corresponding to the t-th initial keyword of the k-th text release information; represents the frequency of the word corresponding to the t-th initial keyword in the k-th text release information, M represents the number of the initial keywords contained in the k-th text release information, represents the frequency of the word corresponding to the mth initial keyword in the kth text release information, lg represents a logarithmic function with base 10, represents the maximum number of texts corresponding to the t-th initial keyword corresponding to the k-th text release information, represents the number of the first texts corresponding to the t-th initial keyword contained in all the text release information, Indicates the number corresponding to all the text release information.

[0042] Exemplarily, interference factors in the text release information, such as HTML tags, special symbols, punctuation marks, etc., are removed to ensure the purity of the text to obtain the preprocessed text, and then the preprocessed text is split into individual words or phrases using the jieba word segmentation, and those words that frequently appear in the text but have little effect on expressing the core meaning, such as Chinese stop words such as "的", "是", and "在", are removed, so as to obtain the initial keywords corresponding to the text release information.

[0043] Exemplarily, the number of times each initial keyword appears is counted in each text release information to obtain the word frequency of each initial keyword in the text release information, and then the number of texts containing a certain initial keyword in all text release information is counted. This number is the first text number.

[0044] Exemplarily, the classification model structure is determined from a machine learning classification algorithm such as Naive Bayes, support vector machine, etc. or a deep learning classification algorithm such as a convolutional neural network, and then the classification model structure is trained using the labeled text data and the text type corresponding to the text data, thereby obtaining a target classification model. The corresponding text type when the target classification model distinguishes types can be defined according to needs. For example, the text type can be a type distinguished based on different target locations, or a type distinguished based on different user behaviors.

[0045] Exemplarily, the trained target classification model is used to classify the text type of the text release information so as to obtain the text type corresponding to the text release information.

[0046] Exemplarily, the statistical text types obtain all the text types included in all the text release information, and then obtain the text release information included under each text type, so as to view the number of texts containing the initial keyword under each text type, that is, obtain the second text quantity, and then obtain the maximum value corresponding to the second text quantity, so as to obtain the maximum text quantity corresponding to the initial keyword under the text type.

[0047] Exemplarily, comprehensively considering the word frequency, the first text quantity, and the maximum text quantity, and then combining the following formula to obtain the word importance:

[0048]

[0049] Wherein, represents the word importance corresponding to the t-th initial keyword in the k-th text release information; represents the word frequency corresponding to the t-th initial keyword included in the k-th text release information, M represents the number of initial keywords included in the k-th text release information, represents the word frequency corresponding to the m-th initial keyword included in the k-th text release information, lg represents the logarithmic function with base 10, represents the maximum text quantity corresponding to the t-th initial keyword corresponding to the k-th text release information, represents the first text quantity corresponding to the t-th initial keyword included in all the text release information, represents the quantity corresponding to all the text release information.

[0050] Exemplarily, in the above formula, it is considered that if the initial keyword appears frequently in a certain text type, according to expert experience, it can be considered that this word can better represent this text type. However, in the calculation process of word importance in the prior art, no matter how many times it appears in each other article, as long as it appears in more text release information, the word importance of this initial keyword will be lower, and thus it is impossible to accurately determine the word importance of this initial keyword. Therefore, the present application adjusts the word importance of the initial keyword by introducing the maximum text quantity corresponding to the initial keyword under the text type, so that the high-frequency words that appear frequently in the text type and can well represent this text type will not be judged to have a low importance due to appearing in multiple articles. That is, this method introduces the factor of the maximum text quantity to adjust the word importance of the initial keyword, which can correct this deviation and more accurately measure the importance of each initial keyword to the text.

[0051] For example, an importance threshold is set, the initial keywords are sorted from high to low according to the importance of the words, and the keywords with importance higher than the threshold are selected as target keywords. These target keywords can more accurately represent the core content of the text release information.

[0052] Exemplarily, clustering algorithms such as K-means clustering, hierarchical clustering, etc. are used to perform text clustering according to target keywords corresponding to the text release information to obtain target clustering results corresponding to the text release information.

[0053] Step S102: Perform feature analysis on each sub-class cluster in the target clustering result to obtain corresponding relevant feature information.

[0054] Exemplarily, keyword extraction is performed on the text release information involved in each sub-cluster in the target clustering result to obtain keyword information corresponding to the text release information involved in each sub-cluster, and then the keyword information is unioned to obtain relevant feature information corresponding to each sub-cluster.

[0055] Step S103: performing topic identification on the target clustering result according to the relevant feature information to obtain an initial topic and relevant text information corresponding to the initial topic.

[0056] Exemplarily, the frequency of occurrence of each sub-keyword in the text of the sub-class cluster in the relevant feature information, the distribution of the sub-keyword in the entire target clustering result, etc. are obtained, and then the importance is calculated according to the frequency of occurrence of the sub-keyword in the text of the sub-class cluster and the distribution in the entire target clustering result to obtain the keyword importance corresponding to each sub-keyword in the relevant feature information. For example, if a sub-keyword appears frequently in the text of a sub-class cluster and rarely appears in other sub-class clusters, then its importance to the sub-class cluster may be high.

[0057] Exemplarily, the importance of keywords is determined as the main sorting basis, and the sub-keywords in the relevant feature information are arranged in order from high to low. According to the constructed sorting rules, all sub-keywords are sorted to form an ordered keyword list. According to actual needs and specific tasks, the number of top-ranked keywords to be selected is determined, and the first k keywords are selected from the sorted keyword list. These keywords represent the most important and representative features in each sub-cluster.

[0058] For example, the k selected keywords are combined in different ways. A single keyword can be used directly as an initial topic, or a combination of two or more keywords can form a more complex initial topic. The generated initial topic is evaluated to determine whether it is practical and representative. Those combinations that are semantically ambiguous, unreasonable, or irrelevant to the sub-cluster theme are removed.

[0059] Exemplarily, it is determined whether the text is closely related to the initial topic by checking whether each text in the subclass cluster contains the keywords in the initial topic, as well as the occurrence frequency and context relationship of the keywords, so as to obtain the relevant text information corresponding to each initial topic.

[0060] In some embodiments, the relevant feature information is the relevant keywords corresponding to the subclass cluster. The topic recognition of the target clustering result according to the relevant feature information to obtain the initial topic and the relevant text information corresponding to the initial topic includes: obtaining the frequency information of the relevant keywords in each sub-text information in the subclass cluster, and obtaining the text length corresponding to the sub-text information; determining the average length corresponding to the subclass cluster according to the text length, and adjusting the frequency information according to the text length and the average length to obtain the first target frequency of the relevant keywords in the subclass cluster; obtaining the third text quantity corresponding to the subclass cluster when the relevant keywords are included, and determining the word weight corresponding to the relevant keywords by combining the third text quantity and the first target frequency with an exponential function; obtaining the initial word vector corresponding to the relevant keywords, and determining the target feature vector corresponding to the subclass cluster by combining the word weight and the initial word vector; clustering the subclass cluster according to the target feature vector to obtain a cluster clustering result, and determining the initial topic and the relevant text information corresponding to the initial topic according to the cluster clustering result; wherein, the word weight is obtained according to the following formula:

[0061]

[0062] Wherein, represents the word weight corresponding to the d-th relevant keyword in the q-th subclass cluster, represents the quantity corresponding to the sub-text information included in the q-th subclass cluster, represents the frequency information of the d-th relevant keyword included in the j-th sub-text information of the q-th subclass cluster, represents the text length corresponding to the j-th sub-text information of the q-th subclass cluster, and ln represents the logarithmic function with the natural constant as the base, represents a regulation parameter, represents the average length corresponding to the q-th subclass cluster, represents the third text quantity corresponding to the q-th subclass cluster when the d-th relevant keyword is included, and exp represents the exponential function.

[0063] Exemplarily, traverse each sub - text information in the subclass cluster. For each relevant keyword, count the number of times it appears in the sub - text information and record it as the frequency information of the keyword in the sub - text. At the same time, calculate the text length of each sub - text information. The text length can be the number of characters, words, etc. in the sub - text, and select an appropriate measurement method according to specific requirements.

[0064] Exemplarily, add up the text lengths of all sub - text information in the subclass cluster, and then divide by the number of sub - texts to obtain the average length corresponding to the subclass cluster. For the frequency information of each relevant keyword in each sub - text, adjust it according to the relationship between the text length of the sub - text and the average length of the subclass cluster. If the text length of the sub - text is significantly longer than the average length, it may mean that the frequency of the keyword appearing in the long text will naturally be higher. At this time, the frequency of the keyword can be appropriately reduced; conversely, if the text length of the sub - text is shorter than the average length and the keyword appears relatively frequently, the frequency of the keyword can be appropriately increased. The adjusted frequency is the first target frequency corresponding to the relevant keyword in the subclass cluster. For example, the first target frequency is obtained according to the following formula:

[0065]

[0066] Where, represents the first target frequency corresponding to the d - th relevant keyword in the j - th sub - text information of the q - th subclass cluster, represents the frequency information of the d - th relevant keyword in the j - th sub - text information of the q - th subclass cluster, represents the text length corresponding to the j - th sub - text information of the q - th subclass cluster, ln represents the natural logarithm function with the natural constant as the base, represents the adjustment parameter, represents the average length corresponding to the q - th subclass cluster.

[0067] Exemplarily, count the number of sub - texts in the subclass cluster that contain the relevant keyword, denoted as the third text quantity. Determine the word weight corresponding to the relevant keyword by combining the third text quantity and the target frequency using the exponential function according to the following formula;

[0068]

[0069] Where, represents the word weight corresponding to the d - th relevant keyword in the q - th subclass cluster, represents the quantity corresponding to the sub - text information in the q - th subclass cluster, represents the frequency information of the d - th relevant keyword in the j - th sub - text information of the q - th subclass cluster, denotes the text length corresponding to the j-th sub-text information of the q-th sub-class cluster, and ln denotes the logarithmic function with the natural constant as the base. denotes the adjustment parameter. denotes the average length corresponding to the q-th sub-class cluster. denotes the third text quantity when the q-th sub-class cluster contains the d-th relevant keyword, and exp denotes the exponential function.

[0070] Exemplarily, for each relevant keyword, its corresponding initial word vector is obtained through a pre-trained word vector model (such as Word2Vec, GloVe, etc.). Then, the initial word vector of each relevant keyword is multiplied by its corresponding word weight. Next, the weighted vectors of all relevant keywords are combined (such as summing or averaging) to obtain the target feature vector corresponding to the sub-class cluster. This target feature vector synthesizes the importance and semantic information of the relevant keywords.

[0071] Exemplarily, a suitable clustering algorithm such as K Means clustering, hierarchical clustering, etc. is used to cluster the target feature vectors of all sub-class clusters. The clustering algorithm will divide the sub-class clusters into different clusters according to the similarity between the target feature vectors, and after clustering, the cluster clustering result is obtained.

[0072] Exemplarily, the sub-class clusters involved in each sub-clustering result in the cluster clustering result are obtained. Then, the relevant keywords corresponding to each sub-class cluster are obtained, and intersection processing is performed according to the relevant keywords to obtain the same keywords. Then, the initial topic is determined according to the relevant keywords. Thus, the text information of the sub-class clusters involved in each sub-clustering result is determined as the relevant text information corresponding to the initial topic.

[0073] Step S104: Perform association analysis based on the relevant text information and the target venue to obtain the target topic corresponding to the target venue and the associated text information corresponding to the target topic.

[0074] Exemplarily, location recognition is performed on each sub-information in the relevant text information corresponding to the initial topic, and then the topic occurrence locations involved in each sub-information are obtained. Then, data statistics are performed according to the topic occurrence locations to obtain the text quantity associated with each topic occurrence location involved in the initial topic. Next, an association analysis is performed between the topic occurrence location and the target venue. When the topic occurrence location is the same as or close to the target venue, it is judged whether the text quantity associated with the topic occurrence location is greater than the preset quantity. When the text quantity associated with the topic occurrence location is greater than the preset quantity, the initial topic is determined as the target topic corresponding to the target venue, and when the topic occurrence location of each sub-information in the relevant text information is the same as or close to the target venue, the sub-information is determined as the associated text information corresponding to the target topic.

[0075] In some embodiments, the obtaining of the target topic corresponding to the target venue and the associated text information corresponding to the target topic through the association analysis based on the relevant text information and the target venue includes: obtaining the associated keywords corresponding to the relevant text information from the cluster clustering result, and pairwise combining the associated keywords to obtain keyword groups, where the keyword groups include a first keyword and a second keyword; retrieving in the relevant text information to obtain a second target frequency corresponding to the simultaneous presence of the first keyword and the second keyword; establishing an initial word association network corresponding to the relevant text information based on the second target frequency in combination with the first keyword and the second keyword; obtaining the initial position information corresponding to the relevant text information, and establishing a word space network between the associated keywords and the initial position information based on the initial position information and the initial word association network; retrieving the target venue in the word space network to obtain multiple relevant word association networks associated with the target venue; performing network merging on the multiple relevant word association networks to obtain an intermediate word association network corresponding to the target venue; performing network screening on the intermediate word association network to obtain a target word association network corresponding to the target venue; and determining the target topic corresponding to the target venue and the associated text information corresponding to the target topic based on the target word association network.

[0076] Exemplarily, associated sub-clusters are obtained from the cluster clustering result, and then the associated keywords corresponding to each relevant text information are obtained based on the sub-clusters.

[0077] Exemplarily, the extracted associated keywords are pairwise combined to form keyword groups. Each keyword group contains a first keyword and a second keyword. Then, for each keyword group, a comprehensive search is performed in the relevant text information, and the number of times the first keyword and the second keyword appear simultaneously is counted. This number is the second target frequency. The second target frequency determines the association weight between the first keyword and the second keyword. Then, an association is established between the first keyword and the second keyword based on the association weight. The higher the second target frequency, the stronger the association between the two keywords. In this way, an initial word association network is constructed for the relevant text information corresponding to each initial topic. This network shows the association relationship and tightness between the keywords. That is, the first keyword and the second keyword are determined as nodes in the initial word association network, and the association weight is determined as the edge between the node corresponding to the first keyword and the node corresponding to the second keyword in the initial word association network.

[0078] Exemplarily, determine the initial position information corresponding to the relevant text information. The initial position information can be the place where the text is published, the geographical location involved, etc. Associate the associated keywords with the initial position information, and combine with the initial word association network to construct a word space network. The word space network not only reflects the association between keywords, but also reflects the connection between keywords and initial position information such as geographical location. Among them, the format corresponding to the word space network is initial position information - initial word association network, that is, the word space network is the association result of the initial position information corresponding to the relevant text information and the initial word association network corresponding to the relevant text information.

[0079] Exemplarily, with the target venue as the retrieval condition, search in the word space network for multiple relevant word association networks associated with the position information where the initial position information is the same as or similar to the target venue.

[0080] Exemplarily, merge the retrieved multiple relevant word association networks, remove duplicate nodes and edges, and integrate their information to form a more comprehensive intermediate word association network, which contains more keyword association information related to the target venue.

[0081] Exemplarily, screen the intermediate word association network according to certain rules, such as removing nodes and edges with low association degrees, retaining important keywords, etc. After screening, obtain the target word association network corresponding to the target venue, which more concisely and accurately reflects the relationship between keywords closely related to the target venue.

[0082] Exemplarily, conduct in-depth analysis on the target word association network, find out the representative keyword combinations among them, and determine the target topic corresponding to the target venue according to these keyword combinations. According to the target topic, screen out the corresponding associated text information from the relevant text information. These text information are closely related to the target topic and can further elaborate on the content of the target topic.

[0083] In some embodiments, obtaining the intermediate word association network corresponding to the target venue after network merging of the multiple relevant word association networks includes: arbitrarily obtaining a first word association network and a second word association network from the multiple relevant word association networks; obtaining the association weight between a third keyword and a fourth keyword in the first word association network; obtaining a first phrase associated with the third keyword from the first word association network, and calculating the first weight between the third keyword and the first phrase; obtaining a second phrase associated with the fourth keyword from the first word association network, and calculating the second weight between the fourth keyword and the second phrase; determining a first network representation value corresponding to the first word association network according to the association weight, the first weight, and the second weight; obtaining a fifth keyword associated with the first word association network from the second word association network; merging the fifth keyword into the first word association network to update the first word association network to obtain a third word association network; recalculating a second network representation value corresponding to the third word association network, and determining the association type between the fifth keyword and the first word association network according to the first network representation value and the second network representation value; performing network merging on the first word association network and the second word association network according to the association type and the fifth keyword to obtain the intermediate word association network corresponding to the target venue; wherein, the first network representation value is obtained according to the following formula:

[0084]

[0085] Wherein, represents the first network representation value corresponding to the h-th first word association network, represents the sum of the association weights between all any two keywords in the h-th first word association network, N represents the number of keywords corresponding to the h-th first word association network, represents the association weight between the a-th third keyword and the b-th fourth keyword corresponding to the h-th first word association network, represents the first weight corresponding to the a-th third keyword corresponding to the h-th first word association network, represents the second weight corresponding to the b-th fourth keyword corresponding to the h-th first word association network, represents a fixed parameter.

[0086] Exemplarily, two networks are randomly selected from multiple related word association networks, named the first word association network and the second word association network respectively. In the first word association network, the third keyword and the fourth keyword are determined, and then the association weight between these two keywords is obtained. This weight is usually determined based on factors such as the frequency of co-occurrence of keywords in the text when constructing the word association network.

[0087] Exemplarily, find the first phrase associated with the third keyword in the first word association network, and then obtain the edge information between the third keyword and each first word in the first phrase. Thus, sum up all the edge information between the third keyword and the first words in the first phrase to obtain the first weight between the third keyword and the first phrase. Similarly, find the second phrase associated with the fourth keyword and calculate the second weight between the fourth keyword and the second phrase.

[0088] Exemplarily, the first network representation value corresponding to the first word association network is determined by the following formula by integrating the association weight, the first weight, and the second weight. The first network representation value can comprehensively reflect the association strength between keywords in the first word association network:

[0089]

[0090] where, represents the first network representation value corresponding to the h-th first word association network, represents the sum of the association weights between all any two keywords in the h-th first word association network, N represents the number of keywords corresponding to the h-th first word association network, represents the association weight between the a-th third keyword and the b-th fourth keyword corresponding to the h-th first word association network, represents the first weight corresponding to the a-th third keyword corresponding to the h-th first word association network, represents the second weight corresponding to the b-th fourth keyword corresponding to the h-th first word association network, represents a fixed parameter.

[0091] Exemplarily, in the second word association network, find the fifth keyword associated with the first word association network, that is, the current keyword in the first word association network also exists in the second word association network. Thus, find the connected keyword corresponding to the current keyword in the second word association network, and this connected keyword does not exist in the first word association network. Then, determine this connected keyword as the fifth keyword associated with the first word association network.

[0092] Exemplarily, the fifth keyword is merged into the first word association network, and the edge information of the connected keywords and the current keyword in the second word association network is also added to the first word association network, so as to obtain the updated third word association network.

[0093] Exemplarily, the second network representation value corresponding to the third word association network is recalculated. The calculation method is similar to that for determining the first network representation value. Then, the first network representation value and the second network representation value are compared, and the association type between the fifth keyword and the first word association network is determined according to the change situation between the two (such as increase, decrease, unchanged, etc.). For example, if the second network representation value increases significantly, it may indicate that the fifth keyword and the first word association network have a strong positive association; if it decreases, it may be a negative association; if the change is small, it may be a weak association.

[0094] Exemplarily, when the association type is a strong association, the fifth keyword in the second word association network is added to the first word association network, and then a new first word association network is obtained. Then, a new fifth keyword is obtained from the second word association network, and it is determined whether the new fifth keyword needs to be merged into the new first word association network until the association type between the words in the second word association network and the first word association network is all negative association or weak association, and finally the intermediate word association network corresponding to the target place is obtained.

[0095] Step S105: Perform behavior detection on the associated text information to obtain the target behavior type corresponding to the target topic.

[0096] Exemplarily, multiple training texts are prepared and behavior type annotations are performed on the training texts. The behavior types include safe, slightly dangerous, moderately dangerous, and highly dangerous. "Safe" can be defined as the behavior or situation described in the text will not cause any harm to personnel, property, or the environment; "slightly dangerous" may refer to the existence of some potential small risks, but it is less likely to cause serious consequences; "moderately dangerous" means there is a certain risk and may cause a certain degree of damage; "highly dangerous" means there is a high risk and is very likely to cause serious casualties, major property losses, or environmental damage.

[0097] Exemplarily, the labeled training texts and their corresponding labeling results are sorted out to form a standardized data set. To evaluate the performance of the model, the data set usually needs to be divided into a training set, a validation set, and a test set. Generally, the training set accounts for most of the data (such as 70%-80%) and is used for training the model; the validation set accounts for 10%-15% and is used to adjust the hyperparameters of the model during the training process; the test set accounts for 10%-15% and is used to finally evaluate the performance of the model.

[0098] Exemplarily, a model structure based on a neural network or a deep learning network is selected. Common models include multi-layer perceptrons, convolutional neural networks, recurrent neural networks and their variants, Transformer architectures, etc. Then, according to the selected model type, a specific model architecture is designed, such as determining parameters such as the number of layers of the model, the number of neurons in each layer, the activation function, the loss function, etc. Thus, parameter training is performed on the designed model architecture according to the training data to obtain a trained text classification model.

[0099] Exemplarily, behavior detection is performed on the associated text information according to the text classification model to obtain a target behavior type corresponding to the target topic, and the target behavior type is one of safe, mildly dangerous, moderately dangerous, and severely dangerous.

[0100] Step S106: When the target behavior type meets the preset type, collect the latest text information associated with the target venue at the second time.

[0101] Exemplarily, the preset types include moderately dangerous and severely dangerous. Then, when the target behavior type is one of the preset types, it is determined that the target behavior type meets the preset type. When the target behavior type is not one of the preset types, it is determined that the target behavior type does not meet the preset type.

[0102] Exemplarily, the second time is the latest time or the time corresponding to after the first time. When the target behavior type meets the preset type, the latest text information associated with the target venue at the second time is collected using web crawler technology. The latest text information is the information related to the target venue published by relevant users at the second time or the information corresponding to the information published by relevant users at the second time when the information location is the target venue.

[0103] Exemplarily, when the target behavior type does not meet the preset type, continue to monitor the information published by users at the target venue.

[0104] Step S107: Perform a popularity analysis on the target topic according to the latest text information and the relevant text information to obtain a target popularity type corresponding to the target topic.

[0105] Exemplarily, the word vector-based method converts the latest text information and relevant text information into word vector representations respectively, and then uses cosine similarity to calculate the cosine value of the angle between the two text vectors. The closer the value is to 1, the higher the similarity, so as to obtain the text similarity between the latest text information and the relevant text information. Then, a suitable similarity threshold is determined according to the actual requirements and data characteristics. For example, if you want to screen out texts that are highly similar to the relevant text information, you can set the threshold relatively high; if you want to expand the screening range, you can set the threshold relatively low, and then screen out the latest text information whose text similarity is greater than or equal to the preset threshold. These screened text information are the similar text information that is close to the relevant text information.

[0106] Exemplarily, for the screened similar text information, extract the release time of each text from the text data. The release time can be a specific date and time, or it can be a date. Find the earliest release time and the latest release time among the similar text information. Subtract the earliest release time from the latest release time to get the time span. Then count the number of similar text information. Divide the number of similar text information by the time span to get a ratio. This ratio is the heat analysis value corresponding to the target topic, which reflects the frequency of text generation related to the target topic within a certain period of time.

[0107] Exemplarily, different heat types are predefined, such as "high heat", "medium heat", "low heat", etc., and a corresponding heat analysis value range is set for each heat type. For example, a heat analysis value greater than 10 is "high heat", a heat analysis value between 3 and 10 is "medium heat", and a heat analysis value less than 3 is "low heat". Thus, according to the calculated heat analysis value, look up the corresponding heat type range in the heat type mapping table, and then determine the target heat type corresponding to the target topic. For example, if the heat analysis value is 15, according to the above mapping table, the target heat type of the target topic is "high heat".

[0108] In some embodiments, obtaining the target heat type corresponding to the target topic by performing heat analysis on the target topic according to the latest text information and the relevant text information includes: obtaining the topic keywords involved in the target topic, and counting the first quantity corresponding to the topic keywords in the relevant text information and the second quantity corresponding to the topic keywords in the latest text information; determining the target quantity corresponding to the topic keywords according to the first quantity and the second quantity; determining the keyword weight corresponding to the topic keywords according to the target quantity, and determining the text feature information corresponding to the latest text information according to the keyword weight and the topic keywords; determining the target relevance between the latest text information and the target topic according to the text feature information; screening out the new text information corresponding to the target topic from the latest text information according to the target relevance; and performing heat analysis on the target topic according to the new text information to obtain the target heat type corresponding to the target topic.

[0109] Exemplarily, a topic model such as the LDA model performs keyword recognition on the associated text information to obtain the topic keywords involved in the target topic, and then traverses the relevant text information to check whether each relevant text contains the topic keywords one by one, and counts the cases where the topic keywords are contained to obtain the total number of times the topic keywords appear in the relevant text information, that is, the first quantity.

[0110] Exemplarily, similarly, traverse the latest text information to check whether each latest text contains the topic keywords. Count the total number of times the topic keywords appear in the latest text information to obtain the second quantity, and then directly add the first quantity and the second quantity to obtain the target quantity, and further calculate the proportion of the target quantity of each topic keyword in the total sum of the target quantities of all topic keywords. This proportion is the keyword weight of the topic keyword.

[0111] Exemplarily, combine the topic keywords and their corresponding keyword weights to form text feature information in the form of a vector or a list. Then, use the cosine similarity method to convert the latest text information into a vector representation with the same dimension as the text feature information, and then calculate the cosine similarity between this vector and the text feature information vector. The similarity value is the target relevance.

[0112] Exemplarily, according to the actual requirements and data characteristics, a suitable correlation threshold is determined. For example, if it is desired to screen out texts highly relevant to the target topic, the threshold can be set higher; if it is intended to expand the screening scope, the threshold can be set lower. The latest text information with a target correlation greater than or equal to the preset threshold is screened out, and this screened text information is the new text information corresponding to the target topic. The text release time corresponding to the new text information is collected, and the time span corresponding to the new text information is calculated based on the text release time. The number of new text information is counted. Dividing the number of new text information by the time span gives a ratio, which can be used as an important indicator for heat analysis.

[0113] Exemplarily, different heat types are predefined, such as "high heat", "medium heat", "low heat", etc., and corresponding heat analysis index ranges are set for each heat type. According to the calculated heat analysis index, the corresponding heat type range is looked up in the heat type mapping table to determine the target heat type corresponding to the target topic.

[0114] Step S108, perform dynamic risk warning on the target venue according to the target heat type and the target behavior type, and obtain the target warning result corresponding to the target venue.

[0115] Exemplarily, when the target heat type is medium heat or high heat, and the target behavior type is medium risk or high risk, then when performing dynamic risk warning on the target venue, the target warning result is that there is a risk in the target venue; when the target heat type is low heat and the target behavior type is low risk, then when performing dynamic risk warning on the target venue, the target warning result is that there is temporarily no risk in the target venue.

[0116] In some embodiments, after obtaining the target warning result, the method further includes: determining the target processing strategy and the relevant processing personnel corresponding to the target venue according to the target behavior type in combination with a preset mapping table; generating a corresponding target description text according to the target behavior type and the target processing strategy; and sending the target description text to the intelligent terminal communicatively connected to the relevant processing personnel, so that the relevant processing personnel obtain the target description text according to the intelligent terminal, and then execute the target processing strategy according to the target description text.

[0117] Exemplarily, a detailed preset mapping table is created in advance, which records the corresponding relationship between different target behavior types and the target handling strategies corresponding to the target places, as well as the relevant handling personnel. When the target early warning result indicates that there is a risk in the target place, the target handling strategy and relevant handling personnel of the target place corresponding to it are searched in the preset mapping table according to the target behavior type. Through this matching, the handling method to be adopted for the current target behavior and the personnel responsible for execution are determined.

[0118] Exemplarily, the key information of the target behavior type and the target handling strategy is integrated to obtain the target description text, and then through the established communication connection, the generated target description text is sent to the intelligent terminals of the relevant handling personnel. The sending method can be text messages, instant messaging software messages, dedicated management application push, etc. Then, after receiving the target description text on the intelligent terminal, the relevant handling personnel carefully read the text content, clarify the target behavior and the target handling strategy to be executed, and then the relevant handling personnel take prompt actions according to the requirements in the target description text and execute the target handling strategy to deal with the situation brought by the target behavior. During the execution process, if necessary, the handling method can be flexibly adjusted according to the actual situation, but it is necessary to ensure that the overall handling direction conforms to the established strategy.

[0119] For example, if a user takes illegal photos on a section with beautiful scenery, the target behavior type is dangerous photography, and the target handling strategy can be prohibiting getting off the vehicle on this section, etc.

[0120] Please refer to Figure 2 , Figure 2An artificial intelligence-driven dynamic risk early warning device 200 for public places provided by an embodiment of the present application. The artificial intelligence-driven dynamic risk early warning device 200 for public places includes a data collection module 201, a feature analysis module 202, a topic recognition module 203, an association analysis module 204, a behavior detection module 205, a data judgment module 206, a heat analysis module 207, and a risk early warning module 208. Among them, the data collection module 201 is used to collect text release information associated with a target place at a first time, and perform text clustering on the text release information to obtain a target clustering result; the feature analysis module 202 is used to perform feature analysis on each sub-cluster in the target clustering result to obtain corresponding relevant feature information; the topic recognition module 203 is used to perform topic recognition on the target clustering result according to the relevant feature information to obtain an initial topic and relevant text information corresponding to the initial topic; the association analysis module 204 is used to perform association analysis on the target place according to the relevant text information to obtain a target topic corresponding to the target place and associated text information corresponding to the target topic; the behavior detection module 205 is used to perform behavior detection on the associated text information to obtain a target behavior type corresponding to the target topic; the data judgment module 206 is used to collect the latest text information associated with the target place at a second time when the target behavior type meets a preset type; the heat analysis module 207 is used to perform heat analysis on the target topic according to the latest text information and the relevant text information to obtain a target heat type corresponding to the target topic; the risk early warning module 208 is used to perform dynamic risk early warning on the target place according to the target heat type and the target behavior type to obtain a target early warning result corresponding to the target place

[0121] In some embodiments, the artificial intelligence-driven dynamic risk early warning device 200 for public places can be applied to a terminal device.

[0122] It should be noted that those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described artificial intelligence-driven dynamic risk early warning device 200 can refer to the corresponding process in the foregoing embodiment of the artificial intelligence-driven dynamic risk early warning method, and will not be described in detail here.

[0123] Please refer to Figure 3 , Figure 3 A schematic block diagram of the structure of a terminal device provided by an embodiment of the present invention.

[0124] As Figure 3As shown, the terminal device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected through a bus 303, and this bus is, for example, an I2C (Inter-integrated Circuit) bus.

[0125] Specifically, the processor 301 is used to provide computing and control capabilities to support the operation of the entire terminal device. The processor 301 can be a central processing unit (CPU), and this processor 301 can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or this processor can also be any conventional processor, etc.

[0126] Specifically, the memory 302 can be a Flash chip, read-only memory (ROM), magnetic disk, optical disc, USB flash drive, or mobile hard disk, etc.

[0127] Those skilled in the art can understand that Figure 3 the structure shown in is only a block diagram of some structures related to the solution of the embodiment of the present invention, and does not constitute a limitation on the terminal device to which the solution of the embodiment of the present invention is applied. The specific server may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0128] Among them, the processor is used to run the computer program stored in the memory and implement any one of the artificial intelligence-driven public place risk dynamic early warning methods provided by the embodiments of the present invention when executing the computer program.

[0129] In one embodiment, the processor is used to run the computer program stored in the memory and implement the following steps when executing the computer program:

[0130] Collect the text release information associated with the target place at the first time, and perform text clustering on the text release information to obtain a target clustering result;

[0131] Perform feature analysis on each sub-cluster in the target clustering result to obtain corresponding relevant feature information;

[0132] Perform topic recognition on the target clustering result according to the relevant feature information to obtain the initial topic and the relevant text information corresponding to the initial topic;

[0133] Perform association analysis on the basis of the relevant text information and the target venue to obtain the target topic corresponding to the target venue and the associated text information corresponding to the target topic;

[0134] Perform behavior detection on the associated text information to obtain the target behavior type corresponding to the target topic;

[0135] When the target behavior type meets the preset type, collect the latest text information associated with the target venue at the second time;

[0136] Perform heat analysis on the target topic according to the latest text information and the relevant text information to obtain the target heat type corresponding to the target topic;

[0137] Perform risk dynamic warning on the target venue according to the target heat type and the target behavior type to obtain the target warning result corresponding to the target venue.

[0138] It should be noted that those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working process of the above-described terminal device can refer to the corresponding process in the embodiment of the method for dynamically warning risks in public places driven by artificial intelligence described above, and will not be elaborated here.

[0139] The embodiment of the present invention also provides a storage medium for computer-readable storage. The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of any of the methods for dynamically warning risks in public places driven by artificial intelligence provided in the specification of the embodiment of the present invention.

[0140] Among them, the storage medium may be an internal storage unit of the terminal device in the foregoing embodiment, such as the hard disk or memory of the terminal device. The storage medium may also be an external storage device of the terminal device, such as a plug-in hard disk equipped on the terminal device, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc.

[0141] Those of ordinary skill in the art will understand that all or some of the steps in the methods disclosed above, and the functional modules / units in systems and devices, can be implemented as software, firmware, hardware, and appropriate combinations thereof. In a hardware embodiment, the division of functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, one physical component can have multiple functions, or one function or step can be executed by several physical components in cooperation. Some or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or can be implemented as hardware, or can be implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include a computer storage medium (or non-transitory medium) and a communication medium (or transitory medium). As is well known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassette, tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, as is well known to those of ordinary skill in the art, a communication medium typically contains computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and can include any information delivery medium.

[0142] It should be understood that the term "and / or" used in the specification and appended claims of the present invention refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. It should be noted that in this article, the term "comprises", "comprising", or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, article, or system comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements that are inherent to such process, method, article, or system. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article, or system comprising that element.

[0143] The serial numbers of the embodiments of the present invention above are only for description and do not represent the superiority or inferiority of the embodiments. The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. An artificial intelligence-driven dynamic risk early warning method for public places, characterized in that, The method includes: Collecting text publishing information associated with a target venue at a first time, and performing text clustering on the text publishing information to obtain a target clustering result; Performing feature analysis on each sub-cluster in the target clustering result to obtain corresponding relevant feature information; Performing topic recognition on the target clustering result according to the relevant feature information to obtain an initial topic and relevant text information corresponding to the initial topic; Performing association analysis on the relevant text information and the target venue to obtain a target topic corresponding to the target venue and associated text information corresponding to the target topic; Performing behavior detection on the associated text information to obtain a target behavior type corresponding to the target topic; When the target behavior type meets a preset type, collecting the latest text information associated with the target venue at a second time; Performing heat analysis on the target topic according to the latest text information and the relevant text information to obtain a target heat type corresponding to the target topic; Performing risk dynamic warning on the target venue according to the target heat type and the target behavior type to obtain a target warning result corresponding to the target venue.

2. The method according to claim 1, characterized in that The performing text clustering on the text publishing information to obtain a target clustering result includes: Performing preprocessing on the text publishing information to obtain initial keywords corresponding to the text publishing information; Obtaining the word frequency corresponding to the initial keywords from the text publishing information, and determining a first text quantity corresponding to when the initial keywords are included from all the text publishing information; Performing text type classification on the text publishing information to obtain a text type corresponding to the text publishing information; Associating the initial keywords with the text type to obtain a second text quantity corresponding to when the initial keywords are included under each text type; Obtaining the maximum value among all the second text quantities to obtain the maximum text quantity, and then determining the word importance corresponding to the initial keywords according to the word frequency in combination with the first text quantity and the maximum text quantity; Screening the initial keywords according to the word importance to obtain target keywords corresponding to the text publishing information; Performing text clustering on the text publishing information according to the target keywords to obtain the target clustering result; Wherein, the word importance is obtained according to the following formula: ； Among them, represents the word importance corresponding to the t-th initial keyword of the k-th text release information; represents the word frequency corresponding to the t-th initial keyword included in the k-th text release information, and M represents the number of initial keywords included in the k-th text release information, represents the word frequency corresponding to the m-th initial keyword included in the k-th text release information, and lg represents the logarithmic function with base 10, represents the maximum number of texts corresponding to the t-th initial keyword of the k-th text release information, represents the first number of texts when the t-th initial keyword is included in all the text release information, represents the number corresponding to all the text release information.

3. The method according to claim 1, characterized in that, The relevant feature information is relevant keywords corresponding to the sub-cluster. The performing topic recognition on the target clustering result according to the relevant feature information to obtain an initial topic and relevant text information corresponding to the initial topic includes: Obtaining the frequency information of the relevant keywords in each sub-text information in the sub-cluster, and obtaining the text length corresponding to the sub-text information; Determining the average length corresponding to the sub-cluster according to the text length, and adjusting the frequency information according to the text length and the average length to obtain a first target frequency corresponding to the relevant keywords in the sub-cluster; Obtain the number of the third texts corresponding to the relevant keywords included in the sub-cluster, and determine the word weight corresponding to the relevant keywords by combining the exponential function with the third text quantity and the first target frequency; Obtain the initial word vector corresponding to the relevant keyword, and determine the target feature vector corresponding to the sub-cluster by combining the word weight with the initial word vector; Perform clustering on the sub-cluster according to the target feature vector to obtain a cluster clustering result, and determine the initial topic and the relevant text information corresponding to the initial topic according to the cluster clustering result; Among them, the word weight is obtained according to the following formula: ； Among them, represents the word weight corresponding to the d-th related keyword in the q-th sub-class cluster, represents the quantity corresponding to the sub-text information contained in the q-th sub-class cluster, represents the frequency information of the d-th related keyword contained in the j-th sub-text information of the q-th sub-class cluster, represents the text length corresponding to the j-th sub-text information of the q-th sub-class cluster, and ln represents the logarithmic function with the natural constant as the base, represents the adjustment parameter, represents the average length corresponding to the q-th sub-class cluster, represents the third text quantity corresponding to the case where the q-th sub-class cluster contains the d-th related keyword, and exp represents the exponential function.

4. The method according to claim 3, wherein The relevant text information and the target venue are associated and analyzed to obtain the target topic corresponding to the target venue and the associated text information corresponding to the target topic, including: Obtain the associated keywords corresponding to the relevant text information from the cluster clustering result, and combine the associated keywords pairwise to obtain keyword groups, where the keyword groups include a first keyword and a second keyword; Retrieve in the relevant text information to obtain the second target frequency corresponding to the simultaneous presence of the first keyword and the second keyword; Establish an initial word association network corresponding to the relevant text information by combining the second target frequency with the first keyword and the second keyword; Obtain the initial position information corresponding to the relevant text information, and establish a word space network between the associated keyword and the initial position information according to the initial position information and the initial word association network; Retrieve in the word space network according to the target venue to obtain multiple relevant word association networks associated with the target venue; After network merging of multiple relevant word association networks, obtain the intermediate word association network corresponding to the target venue; Perform network screening according to the intermediate word association network to obtain the target word association network corresponding to the target venue; Determine the target topic corresponding to the target venue and the associated text information corresponding to the target topic according to the target word association network; 5. The method according to claim 4, characterized in that, The network merging of multiple relevant word association networks to obtain the intermediate word association network corresponding to the target venue includes: Arbitrarily obtain a first word association network and a second word association network from multiple relevant word association networks; Obtain the association weight between the third keyword and the fourth keyword in the first word association network; Obtain the first phrase associated with the third keyword from the first word association network, and calculate the first weight between the third keyword and the first phrase; Obtain the second phrase associated with the fourth keyword from the first word association network, and calculate the second weight between the fourth keyword and the second phrase; Determine the first network representation value corresponding to the first word association network according to the association weight, the first weight, and the second weight; Obtain the fifth keyword associated with the first word association network from the second word association network; Merge the fifth keyword into the first word association network to update the first word association network and obtain a third word association network; Recalculate the second network representation value corresponding to the third word association network, and determine the association type between the fifth keyword and the first word association network according to the first network representation value and the second network representation value; Perform network merging on the first word association network and the second word association network according to the association type and the fifth keyword to obtain the intermediate word association network corresponding to the target venue; Among them, the first network representation value is obtained according to the following formula: ； Among them, represents the first network representation value corresponding to the h-th first word association network, represents the sum of the association weights between all any two keywords in the h-th first word association network, and N represents the number of keywords corresponding to the h-th first word association network, represents the association weight between the a-th third keyword and the b-th fourth keyword corresponding to the h-th first word association network, represents the first weight corresponding to the a-th third keyword corresponding to the h-th first word association network, represents the second weight corresponding to the b-th fourth keyword corresponding to the h-th first word association network, represents a fixed parameter.

6. The method according to claim 1, characterized in that, The method for performing heat analysis on the target topic according to the latest text information and the relevant text information to obtain the target heat type corresponding to the target topic includes: Obtain the topic keywords involved in the target topic, and count the first quantity corresponding to the topic keywords in the relevant text information and the second quantity corresponding to the topic keywords in the latest text information; Determine the target quantity corresponding to the topic keyword according to the first quantity and the second quantity; Determine the keyword weight corresponding to the topic keyword according to the target quantity, and determine the text feature information corresponding to the latest text information according to the keyword weight and the topic keyword; Determine the target association degree between the latest text information and the target topic according to the text feature information; Screen out the new text information corresponding to the target topic from the latest text information according to the target association degree; Perform heat analysis on the target topic according to the new text information to obtain the target heat type corresponding to the target topic.

7. The method according to any one of claims 1-6, characterized in that After obtaining the target warning result, the method further includes: Determine the target processing strategy and relevant processing personnel corresponding to the target venue according to the target behavior type in combination with a preset mapping table; Generate a corresponding target description text according to the target behavior type and the target processing strategy; Send the target description text to an intelligent terminal communicatively connected to the relevant processing personnel, so that the relevant processing personnel can obtain the target description text according to the intelligent terminal, and then execute the target processing strategy according to the target description text.

8. An artificial intelligence-driven dynamic risk early warning device for public places, characterized in that, Including: A data collection module, configured to collect text publishing information associated with a target venue at a first time, and perform text clustering on the text publishing information to obtain a target clustering result; A feature analysis module, configured to perform feature analysis on each sub-cluster in the target clustering result to obtain corresponding relevant feature information; A topic recognition module, configured to perform topic recognition on the target clustering result according to the relevant feature information to obtain an initial topic and relevant text information corresponding to the initial topic; An association analysis module, configured to perform association analysis on the relevant text information and the target venue to obtain a target topic corresponding to the target venue and associated text information corresponding to the target topic; A behavior detection module, configured to perform behavior detection on the associated text information to obtain a target behavior type corresponding to the target topic; A data judgment module, configured to collect the latest text information associated with the target venue at a second time when the target behavior type meets a preset type; A heat analysis module, configured to perform heat analysis on the target topic based on the latest text information and the relevant text information to obtain a target heat type corresponding to the target topic; A risk warning module, configured to perform dynamic risk warning on the target venue according to the target heat type and the target behavior type to obtain a target warning result corresponding to the target venue.

9. A terminal device, characterized in that, The terminal device includes a processor and a memory; The memory is used to store a computer program; The processor is configured to execute the computer program and implement the artificial intelligence-driven dynamic risk warning method for public places according to any one of claims 1 to 7 when executing the computer program.

10. A computer storage medium for computer storage, characterized in that, The computer storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the artificial intelligence-driven dynamic risk warning method for public places according to any one of claims 1 to 7.

Images