News propaganda communication management method, system and program product

Abstract

The invention discloses a news propaganda communication management method and system and a program product. The method comprises the following steps: extracting a subject term set of news text data; secondly, generating a word feature vector and a named entity feature vector of the subject term set, and performing clustering processing on the subject term set based on the word feature vector and the named entity feature vector, so that subject terms with similar word features and named entities are clustered into a class cluster; then, the hot words in the news topic class clusters are recognized by calculating the word popularity of each topic word in each news topic class cluster, so that the hot words in each class cluster are obtained, and the hot word recognition mode based on the hot words and the word popularity can ensure that the pushed hot words have higher attention and news value; therefore, the hot words are generated through a natural language processing technology, and compared with traditional manual topic selection, the time cost is reduced, the efficiency is improved, the subjective experience influence is reduced, and the hit rate of the hot words is guaranteed.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of news information processing, and particularly relates to a news propaganda communication management method, system and program product. BACKGROUND

[0002] With the popularity of the Internet and digital government affairs, information production is growing exponentially, and the government websites, official media and industry platforms at the national, provincial and municipal levels will publish a large amount of news every day. Therefore, the correspondents of each line in news propaganda need to select a propaganda theme from the massive news, so as to create a news propaganda manuscript.

[0003] At present, the correspondents of news propaganda usually manually flip through a large amount of news texts, and rely on qualitative analysis and experience intuition to select topics (i.e., filter out news hot words) for news propaganda. Thus, the existing topic selection for news propaganda has the following deficiencies: (1) the correspondents cannot effectively read, filter and digest the information of a large amount, even if they spend a lot of time every day, they can only touch the tip of the iceberg of information, and it is easy to miss the real hot words with value, so the manual topic selection has the problems of high time cost, low efficiency and difficult topic selection; (2) the topic selection depends too much on experience intuition, which is highly subjective and is likely to miss the real hot words. Therefore, based on the foregoing deficiencies, how to provide a news propaganda communication management method with high efficiency and the ability to push real hot words has become a problem to be solved. SUMMARY

[0004] The purpose of the present application is to provide a news propaganda communication management method, system and program product to solve the problems of low efficiency, high time cost and high subjectivity that cannot hit the real hot words in the prior art.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions: In a first aspect, a news propaganda communication management method is provided, comprising: obtaining all news text data of each news platform within a preset time window; performing topic word identification processing on each news text data to obtain a topic word set corresponding to each news text data; performing named entity recognition on each topic word set to obtain a named entity feature vector of each topic word set; constructing a word feature vector of each topic word set, and performing clustering processing on each topic word set by using the word feature vector and the named entity feature vector of each topic word set to obtain a plurality of news theme clusters; Calculate the word popularity of each topic word in each news topic cluster, and determine the hot words of each news topic cluster based on the word popularity of each topic word in each news topic cluster; Push trending keywords from various news topic clusters to the news dissemination platform to complete the push of trending keywords for news dissemination.

[0006] Based on the aforementioned disclosure, this invention first acquires all news text data from various news platforms within a preset time window; then, it performs keyword identification on each news text data to obtain a keyword set corresponding to each news text data; subsequently, it performs named entity identification on each keyword set to obtain the corresponding named entity feature vector; next, it constructs word feature vectors for each keyword set, and based on the word feature vectors and named entity feature vectors of each keyword set, it performs clustering processing on each keyword set to obtain multiple news topic clusters; thus, this invention essentially performs keyword clustering on news text data from both word features and named entities, thereby clustering news topics with similar word features and named entities into a cluster; finally, this invention identifies the real hot words in each news topic cluster by calculating the word popularity of each keyword in each news topic cluster, and pushes them to the news promotion end, thus completing the hot word push for news promotion.

[0007] Through the above design, this invention extracts a set of topic words from news text data; then, it generates word feature vectors and named entity feature vectors for the topic word set, and performs clustering processing on the topic word set based on these vectors, thereby clustering topic words with similar word features and named entities into a single cluster. This allows for the clustering of topics from massive amounts of news text, providing a topic word library for subsequent hot word identification. Subsequently, this invention calculates the word popularity of each topic word in each news topic cluster to identify hot words within that cluster, thus deriving the hot words in each cluster. Based on this word popularity-based hot word identification method, it ensures that the pushed hot words have higher attention and news value. Therefore, this invention uses natural language processing technology to generate hot words, reducing time costs, improving efficiency, and minimizing the influence of subjective experience compared to traditional manual topic selection, ensuring a high hit rate for hot words. Therefore, this invention is highly suitable for large-scale application and promotion.

[0008] In one possible design, keyword identification processing is performed on each news text data to obtain a set of keywords corresponding to each news text data, including: The preset time window is divided into multiple sampling windows, and each news text data is divided into different sampling windows to obtain the news text data in each sampling window; For any news text data within the k-th sampling window, perform word segmentation on any news text data within the k-th sampling window to obtain a word segmentation set, where the initial value of k is 1; Obtain several consecutive sampling windows preceding the k-th sampling window as the target window; Based on the news text data within the target window, the topic weight of each word in the word segmentation set is calculated; Based on the topic weight of each word, several words are selected from the word segmentation set to form the topic word set of any news text data. After all the news text data in the kth sampling window have been queried, the topic word set of each news text data in the kth sampling window is obtained. Increment k by 1 and re-segment any news text data in the k-th sampling window until k equals K, thus obtaining the topic word set corresponding to each news text data, where K is the total number of sampling windows.

[0009] In one possible design, the topic weight of each word in the word segmentation set is calculated based on the news text data within the target window, including: For any word in the word segmentation set, the first frequency of occurrence of the word in all news text data within the k-th sampling window and the second frequency of occurrence in all news text data within each target window are calculated. Based on the frequency of each second occurrence, the attention characteristics of any of the above words are calculated; Based on each second occurrence frequency and the first occurrence frequency, the word frequency change characteristics of any word are calculated; Using the first occurrence frequency, the attention feature, and the word frequency change feature, the topic weight of any word is calculated, and after all words in the word segmentation set have been queried, the topic weight of each word is obtained.

[0010] In one possible design, the word feature vectors and named entity feature vectors of each topic word set are used to cluster the various topic word sets to obtain multiple news topic clusters, including: For the j-th keyword set, obtain the clustering addition time of the j-th keyword set; Obtain the word feature vectors, named entity feature vectors, and most recent update time for each historical theme cluster in the historical cluster set; Based on the clustering addition time and the most recent update time of each historical topic cluster, invalid topic clusters are deleted from the historical cluster set, so that the remaining historical topic clusters are regarded as valid topic clusters; Based on the word feature vector and named entity feature vector of the j-th topic word set, and the word feature vector and named entity feature vector of each effective topic cluster, the insertion cluster of the j-th topic word set is determined from each effective topic cluster; Add the j-th keyword set to the insertion cluster to obtain the initial news topic cluster; The inserted clusters in the effective cluster set are replaced with the initial news topic clusters to obtain a new effective cluster set, wherein the effective cluster set contains all effective topic clusters; The historical cluster set is updated to a new valid cluster set, and j is incremented by 1. The clustering addition time of the j-th topic word set is re-acquired until j equals m, resulting in several initial news topic clusters. The initial value of j is 1, and m is the total number of topic word sets. Several initial news topic clusters are merged to obtain multiple news topic clusters after merging.

[0011] In one possible design, based on the word feature vector and named entity feature vector of the j-th topic word set, and the word feature vector and named entity feature vector of each effective topic cluster, the insertion cluster of the j-th topic word set is determined from each effective topic cluster, including: Based on the word feature vector and named entity feature vector of the j-th topic word set, and the word feature vector and named entity feature vector of the ith effective topic cluster, the text similarity and named entity similarity between the j-th topic word set and the ith effective topic cluster are calculated. Determine whether the i-th valid topic cluster satisfies the clustering insertion condition, wherein the clustering insertion condition is that the text similarity is greater than the maximum text similarity, or the named entity similarity is greater than the entity similarity threshold; If not, increment i by 1 and recalculate the text similarity and named entity similarity between the j-th topic word set and the i-th effective topic cluster until the i-th effective topic cluster meets the clustering insertion condition. Then, the i-th effective topic cluster that meets the clustering insertion condition is taken as the inserted cluster.

[0012] In one possible design, if all valid topic clusters have been polled, and the last valid topic cluster still does not meet the clustering insertion condition, then the method further includes: Obtain the text similarity and named entity similarity between the j-th topic term set and each valid topic cluster; Determine whether the similarity of each text is less than the minimum text similarity, and whether the similarity of each named entity is less than the entity similarity threshold; If not, sum the text similarity and named entity similarity between the j-th topic word set and any valid topic cluster to obtain the total similarity. After iterating through all valid topic clusters, several total similarities are obtained. From several total similarity scores, the effective topic cluster corresponding to the highest total similarity score is selected as the insertion cluster.

[0013] In one possible design, several initial news topic clusters are merged to obtain multiple news topic clusters after the merging process, including: For any initial news topic cluster, calculate the word popularity of each topic word in that initial news topic cluster; Based on the word popularity of each topic word in any initial news topic cluster, the total popularity of any initial news topic cluster is determined, and after all initial news topic clusters have been queried, the total popularity of each initial news topic cluster is obtained. From a number of initial news topic clusters, remove those whose total popularity is less than the popularity threshold, and use the remaining initial news topic clusters to form a pre-selected topic cluster set; Merge similar pre-selected topic clusters in the pre-selected topic cluster set to obtain multiple news topic clusters after merging.

[0014] In one possible design, the word popularity of each topic term in any initial news topic cluster is calculated, including: For any keyword in any initial news topic cluster, count the first occurrence of the keyword in the initial news topic cluster and the second occurrence of the keyword in all news text data; Obtain the average word count of the news text data corresponding to all keyword sets contained in any initial news topic cluster; From all news text data, select news text data containing any of the aforementioned keywords as target news; Obtain the number of reposts, views, and comments for each target news item, and calculate the popularity of each target news item based on these numbers. The word popularity of any given topic is calculated based on the popularity of each target news item, the number of first occurrences, the number of second occurrences, and the average number of words.

[0015] Secondly, a news and publicity communication management system is provided, including: The acquisition unit is used to acquire all news text data from various news platforms within a preset time window. The topic word recognition unit is used to perform topic word recognition processing on each news text data to obtain the topic word set corresponding to each news text data; The named entity recognition unit is used to perform named entity recognition on each topic word set to obtain the named entity feature vector of each topic word set; Clustering units are used to construct word feature vectors for each topic word set, and to use the word feature vectors and named entity feature vectors of each topic word set to perform clustering processing on each topic word set to obtain multiple news topic clusters; The hot word recognition unit is used to calculate the word popularity of each topic word in each news topic cluster, and to determine the hot words of each news topic cluster based on the word popularity of each topic word in each news topic cluster. The push unit is used to push trending keywords from various news topic clusters to the news dissemination platform to complete the push of trending keywords for news dissemination.

[0016] Thirdly, a news publicity and communication management device is provided. Taking the device as an electronic device as an example, it includes a memory, a processor, and a transceiver that are connected in sequence. The memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the news publicity and communication management method as described in the first aspect or any possible design of the first aspect.

[0017] Fourthly, a storage medium is provided, on which instructions are stored, which, when executed on a computer, perform the news and publicity communication management method as described in the first aspect or any possible design of the first aspect.

[0018] Fifthly, a computer program product containing instructions is provided, which, when executed on a computer, cause the computer to perform the news and publicity communication management method as described in the first aspect or any possible design of the first aspect.

[0019] Beneficial effects: (1) This invention generates hot words through natural language processing technology. Compared with traditional manual topic selection, it reduces time costs, improves efficiency, reduces the influence of subjective experience, and ensures the hit rate of hot words. Therefore, this invention is very suitable for large-scale application and promotion. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the steps of the news publicity and communication management method provided in an embodiment of the present invention. Figure 2 A schematic diagram of the structure of the news publicity and communication management system provided in an embodiment of the present invention; Figure 3This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0022] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0023] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0024] Example: See Figure 1As shown, the news and publicity communication management method provided in this embodiment extracts a set of topic words from news text data; then, it generates word feature vectors and named entity feature vectors for the topic word set, and performs clustering processing on the topic word set based on the word feature vectors and named entity feature vectors, thereby clustering topic words with similar word features and named entities into a cluster; in this way, the topic clustering of massive news texts can be completed, thus providing a topic word library for subsequent hot word identification; then, this method calculates the word popularity of each topic word in each news topic cluster to identify hot words in the news topic cluster, thereby obtaining hot words in each cluster, and based on this, based on words The hot topic keyword identification method ensures that the pushed hot topics have higher attention and news value. Therefore, this method generates hot topics through natural language processing technology, which reduces time costs, improves efficiency, and reduces the influence of subjective experience compared to traditional manual topic selection, thus ensuring the hit rate of hot topics. Therefore, this method is very suitable for large-scale application and promotion. For example, this method can be run on the news information processing terminal. Optionally, the news information processing terminal can be, but is not limited to, a server or a computer. It is understood that the aforementioned execution entity does not constitute a limitation on the embodiments of this application. Accordingly, the operation steps of this method can be, but are not limited to, the steps S1 to S6 below.

[0025] S1. Obtain all news text data from various news platforms within a preset time window; in this embodiment, for example, but not limited to, obtaining news text data from platforms corresponding to the National Bureau, Provincial Bureau, and Municipal Bureau within a preset time window; wherein, the preset time window can be 1 hour, 2 hours, 6 hours, or 12 hours before the current time; of course, the specific time window for collecting news text data can be set according to actual use, which will not be elaborated here.

[0026] After crawling the news text data of various news platforms within a preset time window, keyword recognition can be performed to obtain the keyword set of each news text data. The process is shown in step S2 below.

[0027] S2. Perform keyword recognition processing on each news text data to obtain the keyword set corresponding to each news text data; in specific implementation, this embodiment divides the preset time window into multiple sampling windows, and then divides each news text data into different sampling windows; then, based on the news text data in the sampling window and the previous sampling windows, keyword recognition is performed on each news text data in the sampling window, and the process is as shown in the following steps S21 to S26.

[0028] S21. Divide the preset time window into multiple sampling windows, and assign each news text data to a different sampling window to obtain the news text data within each sampling window. In specific implementation, the preset time window is divided starting from the beginning time. For example, assuming the sampling window length is 10 minutes and the preset time window is from 12:00 to 1:00, then 12:00 to 12:10 is the first sampling window, 12:10 to 12:20 is the second sampling window, and so on, thus dividing the preset time window into 6 sampling windows. Of course, the above example is only an example, and this embodiment is not limited to this.

[0029] After the preset time window is divided, the news text data can be divided. For any sampling window, for example, but not limited to, news text data whose news release time is within any sampling window can be divided into the sampling window. In this way, after all sampling windows have been polled, the news text data in each sampling window can be determined.

[0030] After determining the news text data in each sampling window, keyword identification can be performed on each news text data in each sampling window, as shown in steps S22 to S26 below.

[0031] S22. For any news text data within the k-th sampling window, perform word segmentation on any news text data within the k-th sampling window to obtain a word segmentation set, where the initial value of k is 1; in this embodiment, for example, but not limited to, stop word removal can be performed first, and then the jieba word segmentation tool can be used to perform word segmentation on any news text data to obtain the corresponding word segmentation set.

[0032] After obtaining the word segmentation set, the topic weight of each word in the word segmentation set can be calculated by combining the news text data in several sampling windows before the k-th sampling window. The process is shown in steps S23 and S24 below.

[0033] S23. Obtain several consecutive sampling windows before the k-th sampling window as target windows; in this embodiment, the (k-1), (k-2), (k-3), (k-4), and (k-5)-th sampling windows can be obtained as target windows, that is, the five sampling windows before the k-th sampling window are obtained as target windows; at the same time, when k is 1, the last five sampling windows in the previous preset time window are taken as target windows.

[0034] After obtaining the target window, the topic weights of words in the word segmentation set can be calculated based on the news text data within the target window, as shown in step S24 below.

[0035] S24. Calculate the topic weight of each word in the word segmentation set based on the news text data in the target window. In specific applications, this embodiment measures the topic weight of words by using the news text data in the target window and considering three aspects: word frequency, word frequency change, and word attention. The process can be, but is not limited to, the steps S24a to S24d below.

[0036] S24a. For any word in the word segmentation set, calculate the first frequency of occurrence of the word in all news text data within the k-th sampling window, and the second frequency of occurrence of the word in all news text data within each target window. In this embodiment, assuming there are 10 news text data within the k-th sampling window, the first frequency of occurrence of the word is calculated by counting the number of times the word appears in these 10 news text data. Similarly, the method for calculating the second frequency of occurrence of the word in the news text data within each target window is the same, and will not be repeated here.

[0037] After obtaining the frequency of occurrence of any word in the k-th sampling window and each target window, the attention feature of the word can be calculated, as shown in step S24b below.

[0038] S24b. Calculate the attention characteristics of any word based on the frequency of each second occurrence; in specific implementation, for example, but not limited to, the following steps can be used to calculate the attention characteristics of any word.

[0039] S24b1. Based on each second occurrence frequency and the first occurrence frequency, calculate the second average occurrence frequency of any word; in this embodiment, sum all second occurrence frequencies and first occurrence frequencies, and then divide the sum by N+1 to obtain the second average occurrence frequency; where N is the total number of target windows.

[0040] After obtaining the second average frequency of occurrence of any word, the first attention factor can be calculated, as shown in steps S24b2 and S24b3 below.

[0041] S24b2. Calculate the first difference between each second occurrence frequency and the second average occurrence frequency, and the second difference between the first occurrence frequency and the second average occurrence frequency.

[0042] After calculating the difference between each second occurrence frequency and the second average occurrence frequency, and the difference between the first occurrence frequency and the second average occurrence frequency, the first attention factor is calculated based on this, as shown in step S24b3 below.

[0043] S24b3. Calculate the first attention factor based on each of the first differences of the second difference. In specific implementation, for example, but not limited to, first calculate the square of each of the first differences and sum them to obtain the first sum of squares; then, calculate the square of the second difference to obtain the second sum of squares; next, sum the first and second sums of squares to obtain the total sum of squares; finally, take the square root of the total sum of squares to obtain the first attention factor. Thus, the first attention factor is calculated based on the difference between the frequency of any word in each window and its corresponding average frequency of occurrence, which can reflect the global fluctuation of the word frequency of any word in several consecutive windows.

[0044] After obtaining the first attention factor, the second attention factor can be calculated, as shown in step S24b4 below.

[0045] S24b4. Calculate the second attention factor using each second occurrence frequency and the first occurrence frequency; in specific applications, it is possible, but not limited to, first summing each second occurrence frequency to obtain the first average occurrence frequency; then, add 1 to the first occurrence frequency and the first average occurrence frequency respectively to obtain the first frequency sum and the second frequency sum; finally, use the ratio between the first frequency sum and the second frequency sum as the second attention factor; based on this, the second attention factor is used to characterize the continuous fluctuation characteristics of any word between several historical sampling windows and the current sampling window.

[0046] After obtaining the second attention factor, the third attention factor can be calculated, as shown in steps S24b5 and S24b6 below.

[0047] S24b5. From each target window, select the previous target window of the kth sampling window, and take the second occurrence frequency of any word in the selected target window as the target frequency.

[0048] S24b6. Calculate the third attention factor based on the target frequency and the first occurrence frequency.

[0049] In this embodiment, the target frequency and the first occurrence frequency are each added by 1, and then the ratio of the two is taken to obtain the third attention factor; thus, the third attention factor reflects the local fluctuation characteristics of any word between the current sampling window and the previous sampling window.

[0050] Based on this, after obtaining the aforementioned three attention factors, a weighted sum can be performed to obtain the attention feature of any word, as shown in step S24b7 below.

[0051] S24b7. The first attention factor, the second attention factor, and the third attention factor are weighted and summed to obtain the attention feature of any one of the words.

[0052] Therefore, through the aforementioned steps S24b and its sub-steps, this embodiment uses the overall, continuous, and local fluctuation characteristics of words within multiple sampling windows to measure the attention of any word. That is, the more drastic the fluctuation, the greater the attention characteristic, indicating that the attention of any word changes more drastically, and that any word may represent a newsworthy event, and the greater the probability that it is a topic word.

[0053] After calculating the attention level of any word, the word frequency change characteristics can be calculated, as shown in step S24c below.

[0054] S24c. Based on each second occurrence frequency and the first occurrence frequency, calculate the word frequency change feature of any word; in this embodiment, the word frequency change feature mainly refers to the word frequency growth rate; considering that the length of the sampling window is fixed, the word frequency in the current sampling window can be used to represent the growth rate; at the same time, in order to better smooth abrupt word frequencies, this embodiment uses the average word frequency in several sampling windows before the current sampling window to compare with the word frequency in the current sampling window, and uses the difference between the two as the word frequency growth rate; based on this, the calculation process of the word frequency change feature is as follows: first, based on each second occurrence frequency, calculate the first average occurrence frequency of any word in all target windows; then, calculate the difference between the first occurrence frequency and the first average occurrence frequency as the word frequency change feature.

[0055] After obtaining the word frequency growth rate (i.e. word frequency change characteristics), the topic weight of any word can be calculated by combining it with the aforementioned attention characteristics, as shown in step S24d below.

[0056] S24d. Using the first occurrence frequency, the attention feature, and the word frequency change feature, calculate the topic weight of any word, and after polling all words in the word segmentation set, obtain the topic weight of each word; in this embodiment, for example, but not limited to, performing logarithmic operations on the first occurrence frequency, the attention feature, and the word frequency change feature respectively to obtain the first logarithmic operation result, the second logarithmic operation result, and the third logarithmic operation result (assuming the first occurrence frequency is Ck, its logarithmic operation is log(Ck); of course, the logarithmic operation process of the attention feature and the word frequency change feature is also the same, and will not be repeated here); then, perform a weighted sum of the first logarithmic operation result, the second logarithmic operation result, and the third logarithmic operation result to obtain the topic weight of any word; of course, the weight can be specifically set according to actual use, and this embodiment does not make specific limitations.

[0057] Through the aforementioned steps S24a to S24d, the topic weight of each word in the word segmentation set can be calculated; then, based on this, the topic words can be selected, as shown in step S25 below.

[0058] S25. Based on the topic weight of each word, select several words from the word segmentation set to form the topic word set of any news text data. After all news text data in the k-th sampling window have been polled, the topic word set of each news text data in the k-th sampling window is obtained. In this embodiment, each word is sorted in descending order of topic weight to obtain a sorting sequence. Then, the first 3 or 4 words in the sorting sequence are taken as the topic words of any news text data, thereby forming its corresponding topic word set. Based on this, the topic word set of the remaining news text data in the k-th sampling window can be determined in the same way as described above.

[0059] Then, using the same principle, the topic words of the news text data in the remaining sampling windows can be extracted, as shown in step S26 below.

[0060] S26. Increment k by 1 and re-segment any news text data in the k-th sampling window until k equals K, thus obtaining the set of keywords corresponding to each news text data, where K is the total number of sampling windows.

[0061] Thus, through the aforementioned steps S21 to S26, the subject words of each news text data can be identified, and the subject word set of each news text data can be obtained; then, named entity recognition can be performed, the process of which is shown in step S3 below.

[0062] S3. Perform named entity recognition on each topic term set to obtain the named entity feature vector for each topic term set. In practical applications, named entities are the basic information units of text. Common named entities include names of people, places, organizations, time, date, currency, etc. Named entities are keywords that embody the content of the text and can serve as an important basis for determining whether a topic term belongs to a certain topic. Therefore, named entity recognition (Named Entity) is crucial for identifying named entities. Named entity recognition (NER) technology is widely used in topic detection and tracking. Examples include, but are not limited to, using the SpacyNER model or the Bi-LSTM-CRF model (a model combining a bidirectional long short-term memory network (Bi-LSTM) and a conditional random field (CRF). Bi-LSTM captures contextual information in the text, while CRF ensures the generated entity label sequence is valid) to perform named entity recognition on a set of topic words, thus obtaining the corresponding named entity recognition results. Simultaneously, this embodiment uses the BIOE annotation method to identify named entities on the set of topic words (a sequence annotation method used for named entity recognition tasks in the field of natural language processing), thereby forming named entity feature vectors. Of course, the aforementioned named entity recognition and identification are common techniques in natural language processing, and their principles will not be elaborated further.

[0063] After obtaining the named entity feature vectors of each topic word set, word feature vectors can be constructed so that the topic word sets can be clustered based on the word feature vectors and named entity feature vectors, thereby clustering news topics with similar word features and named entities into a cluster; wherein the aforementioned process is as shown in step S4 below.

[0064] S4. Construct word feature vectors for each topic word set, and use the word feature vectors and named entity feature vectors of each topic word set to perform clustering processing on each topic word set to obtain multiple news topic clusters; in this embodiment, for example, but not limited to, inputting each topic word set into the word2vec model for word embedding, thereby obtaining the word feature vectors of each topic word set. Thus, the word feature vector of a topic word set contains the word vector of each topic word in the topic word set; after obtaining the word feature vectors of each topic word set, the named entity feature vectors can be combined to perform clustering processing on the topic word sets.

[0065] In this embodiment, the similarity between the word feature vectors of each topic word set and the word feature vectors of historical topic clusters, as well as the similarity between the named entity feature vectors of each topic word set and the word feature vectors of historical topic clusters, is calculated. Then, based on the aforementioned two similarities, the cluster to which each topic word set belongs is determined. At the same time, a cluster timeliness judgment is added during the clustering process to eliminate invalid clusters. Specifically, the clustering process is shown in steps S41 to S48 below.

[0066] S41. For the j-th keyword set, obtain the clustering addition time of the j-th keyword set; in this embodiment, the clustering addition time of the j-th keyword set is the current time. For example, if the clustering process of the j-th keyword set starts at 1:01 AM in 2015, then its clustering addition time is 1:01 AM in 2015; thus, after obtaining the clustering addition time of the current keyword set to be clustered, the historical cluster set can be obtained, and the process is shown in step S42 below.

[0067] S42. Obtain the word feature vector, named entity feature vector, and most recent update time of each historical topic cluster in the historical cluster set; In this embodiment, the historical cluster set is the cluster generated after the last clustering. For example, at the beginning of the last clustering, there were historical cluster set A, historical cluster set B, and historical cluster set C. Then, a topic word set was added to historical cluster set C, and historical cluster set C was updated. Based on this, in this clustering, the historical cluster set includes historical cluster set A, historical cluster set B, and the updated historical cluster set C; At the same time, the word feature vector of any historical topic cluster... The term feature vector is the mean vector of all the topic word sets and their features contained in the historical topic cluster. Its most recent update time is the clustering addition time of the most recently added topic word set in the historical topic cluster. The named entity feature vector of the historical topic cluster is the union of the named entity feature vectors of all the topic word sets contained in the historical topic cluster. Of course, when the same named entity is merged, it is only recorded once. In this way, after obtaining the term feature vector, named entity feature vector and most recent update time of each historical topic cluster, the timeliness of the cluster can be judged. The process is shown in step S43 below.

[0068] S43. Based on the cluster addition time and the most recent update time of each historical topic cluster, delete invalid topic clusters from the historical cluster set, so that the remaining historical topic clusters are regarded as valid topic clusters; in specific applications, the time difference between the cluster addition time and the most recent update time of each historical topic cluster is calculated; then, historical topic clusters with a time difference greater than the time threshold are deleted as invalid topic clusters.

[0069] After obtaining the effective topic clusters, the j-th topic word set can be clustered, that is, the cluster to which the j-th topic word set belongs can be determined from each effective topic cluster, as shown in step S44 below.

[0070] S44. Based on the word feature vector and named entity feature vector of the j-th topic word set, and the word feature vector and named entity feature vector of each effective topic cluster, determine the insertion cluster of the j-th topic word set from each effective topic cluster; in specific implementation, for example, but not limited to, the following steps S44a to S44c can be used to determine the insertion cluster of the j-th topic word set.

[0071] S44a. Based on the word feature vectors and named entity feature vectors of the j-th topic word set, and the word feature vectors and named entity feature vectors of the i-th effective topic cluster, calculate the text similarity and named entity similarity between the j-th topic word set and the i-th effective topic cluster. In practical applications, as previously explained, the word feature vectors contain the word vectors of each topic word in the topic word set. Therefore, first calculate the cosine distance between the b-th word vector in the word feature vectors of the j-th topic word set and each word vector in the word feature vectors of the i-th effective topic cluster; then, take each... The maximum value of the cosine distances is used as the initial text similarity between the b-th word vector and the i-th effective topic cluster. Then, b is incremented by 1, and the cosine distances between the b-th word vector and each word vector in the word feature vector of the i-th effective topic cluster are recalculated until b equals B, resulting in B initial text similarities. Finally, the mean of the B initial text similarities is calculated as the text similarity between the j-th topic word set and the i-th effective topic cluster. Here, the initial value of b is 1, and B is the total number of word vectors in the word feature vector of the j-th topic word set.

[0072] Thus, after obtaining the text similarity, the named entity similarity between the two can be calculated. The process is as follows: calculate the proportion of named entities in the named entity feature vector of the j-th topic word set that appear in the named entity feature vector of the i-th effective topic cluster, and use this proportion as the named entity similarity between the two.

[0073] In this embodiment, the number of times a named entity in the named entity feature vector of the j-th topic term set appears in the named entity feature vector of the i-th valid topic cluster is counted. Then, the number of occurrences is divided by the total number of named entities in the named entity feature vector of the j-th topic term set to obtain the aforementioned ratio.

[0074] After calculating the text similarity and named entity similarity between the j-th topic word set and the ith effective topic cluster, it can be determined whether the ith effective topic cluster belongs to the j-th topic word set. The process is shown in step S44b below.

[0075] S44b. Determine whether the i-th valid topic cluster satisfies the clustering insertion condition, wherein the clustering insertion condition is that the text similarity is greater than the maximum text similarity, or the named entity similarity is greater than the entity similarity threshold.

[0076] In this embodiment, if the text similarity between the j-th topic term set and the i-th valid topic cluster is greater than the maximum text similarity, or the named entity similarity between them is greater than the entity similarity threshold, then the i-th valid topic cluster can be used as the cluster to which the j-th topic term set belongs, i.e., as the insertion cluster. Conversely, if the aforementioned conditions are not met, then the text similarity and named entity similarity between the j-th topic term set and the next valid topic cluster need to be calculated for further judgment. This process is repeated until a valid topic cluster that meets the clustering insertion conditions is found.

[0077] The aforementioned cyclic search process is shown in step S44c below.

[0078] S44c. If not, increment i by 1 and recalculate the text similarity and named entity similarity between the j-th topic word set and the i-th effective topic cluster until the i-th effective topic cluster meets the clustering insertion condition, and then take the i-th effective topic cluster that meets the clustering insertion condition as the inserted cluster.

[0079] The following example illustrates the steps S44a to S44c. Assume there are 5 effective topic clusters. When the text similarity and named entity similarity between the j-th topic set and the 4th effective topic cluster meet the aforementioned conditions, the 4th effective topic cluster is used as the insertion cluster for the j-th topic set.

[0080] Meanwhile, when all valid topic clusters have been polled and the last valid topic cluster still does not meet the aforementioned clustering insertion conditions, another judgment method needs to be adopted, and the process is as shown in steps S44d to S44g below.

[0081] S44d. Obtain the text similarity and named entity similarity between the j-th topic term set and each valid topic cluster.

[0082] After obtaining the text similarity and named entity similarity between the j-th topic word set and each valid topic cluster, a second judgment can be made, as shown in step S44e below.

[0083] S44e. Determine whether the similarity of each text is less than the minimum text similarity and whether the similarity of each named entity is less than the entity similarity threshold. In specific applications, if the text similarity between the j-th topic word set and all valid topic clusters is less than the minimum text similarity and the named entity similarity between it and all valid topic clusters is less than the entity similarity threshold, then a new valid topic cluster is created, and the word feature vector and named entity feature vector of the j-th topic word set are used as the word feature vector (i.e., cluster center) and named entity feature vector of the new valid topic cluster. Then, the new valid topic cluster is added to the following set of valid clusters to obtain a new set of valid clusters. Finally, the next topic word set can be clustered, i.e., step S47 is executed.

[0084] Meanwhile, if the conditions for the aforementioned secondary judgment are not met, it is necessary to redetermine the valid subject cluster to which the j-th keyword set belongs, as shown in steps S44f and S44g below.

[0085] If S44f is not true, then sum the text similarity and named entity similarity between the j-th topic word set and any valid topic cluster to obtain the total similarity. After iterating through all valid topic clusters, several total similarities are obtained.

[0086] S44g. From several total similarities, select the effective topic cluster corresponding to the largest total similarity as the insertion cluster; in specific implementation, the effective topic cluster with the largest sum of text similarity and named entity similarity can be selected as the insertion cluster of the j-th topic word set.

[0087] Thus, through the aforementioned steps S44 and its sub-steps, the insertion cluster of the j-th keyword set can be determined; then, the clustering of the j-th keyword set can be performed, as shown in step S45 below.

[0088] S45. Add the j-th keyword set to the insertion cluster to obtain the initial news topic cluster; in this embodiment, after adding the j-th keyword set to the insertion cluster, the keywords in the initial news topic cluster are not deduplicated, that is, there may be the same keywords in the initial news topic cluster.

[0089] After the clustering of the j-th keyword set is completed, the effective cluster set can be updated, as shown in step S46 below.

[0090] S46. Replace the inserted clusters in the effective cluster set with the initial news topic clusters to obtain a new effective cluster set, wherein the effective cluster set contains all effective topic clusters; in this embodiment, based on the aforementioned example, it is assumed that historical cluster set A, historical cluster set B, and updated historical cluster set C are all effective topic clusters, that is, effective topic clusters include clusters A, B, and C; at the same time, in this clustering, the j-th topic word set is added to the effective topic cluster A, that is, the effective topic cluster A with the j-th topic word set is used as an initial news topic cluster. The topic cluster is A; at this point, the new effective cluster set includes: the initial news topic cluster A, the effective topic cluster B, and the effective topic cluster C; based on this, the word feature vector (i.e., the cluster center) of the initial news topic cluster A is updated, that is, after the j-th topic word set is added, its cluster center changes; specifically, the mean of the word feature vectors of all topic word sets in the initial news topic cluster A is used as its cluster center; similarly, the command entity feature vector is also updated accordingly, and the update method is to merge the command entity feature vectors of all topic word sets within the cluster.

[0091] Thus, after updating the effective cluster set, the historical cluster set can be updated so that the updated historical cluster set can be used to cluster the next keyword set. The process is shown in step S47 below.

[0092] S47. Update the historical cluster set to a new valid cluster set, increment j by 1, and re-acquire the clustering addition time of the j-th topic word set until j equals m, to obtain several initial news topic clusters, where the initial value of j is 1, and m is the total number of topic word sets.

[0093] Thus, by going through the aforementioned steps S41 to S47, the clustering of each keyword set can be completed. Then, in order to improve the accuracy of clustering, this embodiment also performs cluster merging processing, that is, merging similar clusters and removing noisy clusters, the process of which is shown in step S48 below.

[0094] S48. Merge several initial news topic clusters to obtain multiple news topic clusters after merging. In this embodiment, the aforementioned cluster merging mainly involves first removing low-popularity clusters and then merging similar clusters. The process can be, but is not limited to, the steps S48a to S48d below.

[0095] S48a. For any initial news topic cluster, calculate the word popularity of each topic word in that initial news topic cluster; in specific applications, take any topic word in that initial news topic cluster as an example to illustrate the word popularity calculation process, which may be, but is not limited to, the steps S48a1 to S48a5 below.

[0096] S48a1. For any topic word in any initial news topic cluster, count the first occurrence number of the topic word in the initial news topic cluster and the second occurrence number of the topic word in all news text data.

[0097] After counting the number of times each keyword appears in any initial news topic cluster and the number of times it appears in all news text data within a preset time window, the average word count can be calculated, as shown in step S48a2 below.

[0098] S48a2. Obtain the average word count of the news text data corresponding to all topic word sets contained in any initial news topic cluster; in this embodiment, first obtain the word segmentation set of the news text data corresponding to all topic word sets contained in any initial news topic cluster; then, calculate the total number of words in each word segmentation set; finally, divide the total number of words by the number of word segmentation sets to obtain the average word count.

[0099] After obtaining the average number of words, news text data containing any of the subject words can be acquired to calculate the popularity of news text data containing any of the subject words, and based on this, the word popularity of any of the subject words can be measured; wherein, the news text data filtering process is as shown in step S48a3 below.

[0100] S48a3. From all news text data, select news text data containing any of the aforementioned keywords as target news; after selecting news text data containing the aforementioned keywords, the news popularity can be calculated, as shown in step S48a4 below.

[0101] S48a4. Obtain the number of reposts, views, and comments for each target news item, and calculate the popularity of each target news item based on the number of reposts, views, and comments. In this embodiment, the appearance of hot words is always accompanied by high popularity, and popularity is reflected in the number of views, comments, and reposts of the news. Therefore, this embodiment calculates the popularity of each target news item by obtaining the number of reposts, views, and average of each target news item.

[0102] Specifically, for any target news, its popularity H is calculated as follows: H = 1 + α1 × ln(z) + α2 × ln(y) + α3 × ln(p); where z, y, and p represent the number of reposts, views, and comments of the target news, respectively, and α1, α2, and α3 are the repost weight, view weight, and comment weight, respectively.

[0103] Thus, after calculating the popularity of each target news item, the word popularity of any given topic can be calculated by combining the first occurrence count, the second occurrence count, and the average number of words mentioned above, as shown in step S48a5 below.

[0104] S48a5. Calculate the word popularity of any given topic word based on the popularity of each target news item, the number of first occurrences, the number of second occurrences, and the average number of words. In specific implementation, for example, but not limited to, the following formula can be used to calculate the word popularity of any given topic word.

[0105] ; In the formula, This indicates the popularity of any given keyword. These represent the first occurrence count and the second occurrence count, respectively. The average number of words, Let G represent the popularity of the g-th target news item, and let G represent the total number of target news items.

[0106] Thus, through the aforementioned steps S48a1 to S48a5, this embodiment combines word frequency with the number of reposts, views, and comments on the news text to comprehensively measure the popularity of the topic; based on this, the accuracy of word popularity measurement can be improved.

[0107] After obtaining the word popularity of each topic word in any initial news topic cluster, the total popularity of any initial news topic cluster can be calculated based on this, as shown in step S48b below.

[0108] S48b. Based on the word popularity of each topic word in any initial news topic cluster, determine the total popularity of any initial news topic cluster, and obtain the total popularity of each initial news topic cluster after polling all initial news topic clusters; in this embodiment, the total popularity of any initial news topic cluster is obtained by summing the word popularity of each topic word in any initial news topic cluster; after calculating the total popularity of all initial news topic clusters, noise clusters can be removed, as shown in step S48c below.

[0109] S48c. From a number of initial news topic clusters, remove those whose total popularity is less than the popularity threshold, and use the remaining initial news topic clusters to form a pre-selected topic cluster set.

[0110] After removing the initial news topic clusters whose total popularity is less than the popularity threshold, similar clusters can be merged, as shown in step S48d below.

[0111] S48d. Merge similar pre-selected topic clusters in the pre-selected topic cluster set to obtain multiple news topic clusters after merging. In this embodiment, for example, but not limited to, first selecting any pre-selected topic cluster from the pre-selected topic cluster set as the target cluster; then, calculating the cluster similarity between the target cluster and each specified cluster (where each specified cluster is a pre-selected topic cluster excluding the target cluster); then, merging the specified clusters with cluster similarity greater than the similarity threshold into the target cluster to obtain a news topic cluster after merging; next, deleting the target cluster and the specified clusters with cluster similarity greater than the similarity threshold from the pre-selected topic cluster set to obtain a new pre-selected topic cluster set; finally, updating the pre-selected topic cluster set to the new pre-selected topic cluster set, and re-selecting any pre-selected topic cluster from the pre-selected topic cluster set until all pre-selected topic clusters in the pre-selected topic cluster set have been traversed, thus obtaining several news topic clusters.

[0112] In practice, cluster similarity is the similarity between the word feature vector (i.e., cluster center) of the target cluster and the word feature vector of the specified cluster. The calculation process can be found in the aforementioned text similarity, and will not be repeated here.

[0113] Thus, by removing low-popularity clusters and merging similar clusters through the aforementioned steps S48a to S48d, news topic clusters can be obtained; then, hot words can be extracted from each news topic cluster, as shown in step S5 below.

[0114] S5. Calculate the word popularity of each topic word in each news topic cluster, and determine the hot words of each news topic cluster based on the word popularity of each topic word in each news topic cluster. In this embodiment, the calculation process of the word popularity of topic words in each news topic cluster can be referred to the aforementioned steps S48a1 to S48a5 (that is, combining word frequency level, as well as the number of forwards, views and comments of news text to measure the popularity of topic words, thereby improving the accuracy of hot word identification). The principle will not be repeated here.

[0115] After obtaining the word popularity of each topic word in each news topic cluster, hot words in each news topic cluster can be extracted. For any news topic cluster, the topic words in that news topic cluster can be sorted in descending order of word popularity, but are not limited to this. Then, the top 3 or 5 topic words are taken as the hot words of that news topic cluster. After extracting the hot words of the remaining news topic clusters in the above manner, the hot words can be pushed out, as shown in step S6 below.

[0116] S6. Push the hot keywords of each news topic cluster to the news publicity terminal to complete the hot keyword push for news publicity; in this embodiment, pushing hot keywords to the news publicity terminal can enable news correspondents to determine the news publicity topic based on the hot keywords, thereby improving the efficiency of topic selection.

[0117] Therefore, through the news publicity and communication management method described in detail in steps S1 to S6 above, this invention extracts a set of topic words from news text data; then, it generates word feature vectors and named entity feature vectors for the topic word set, and performs clustering processing on the topic word set based on these vectors, thereby clustering topic words with similar word features and named entities into a cluster; in this way, the clustering of topics from massive amounts of news text can be completed, providing a topic word library for subsequent hot word identification; then, this invention calculates the word popularity of each topic word in each news topic cluster to identify hot words in the news topic cluster, thereby obtaining hot words in each cluster. Based on this, the hot word identification method based on word popularity can ensure that the hot words pushed have higher attention and news value; thus, this invention generates hot words through natural language processing technology, which reduces time costs, improves efficiency, and reduces the influence of subjective experience compared to traditional manual topic selection, ensuring the hit rate of hot words; therefore, this invention is very suitable for large-scale application and promotion.

[0118] like Figure 2 As shown, the second aspect of this embodiment provides a hardware system for implementing the news publicity and communication management method described in the first aspect of the embodiment, including: The acquisition unit is used to acquire all news text data from various news platforms within a preset time window.

[0119] The topic word recognition unit is used to perform topic word recognition processing on each news text data to obtain the topic word set corresponding to each news text data.

[0120] The named entity recognition unit is used to perform named entity recognition on each topic word set to obtain the named entity feature vector of each topic word set.

[0121] The clustering unit is used to construct word feature vectors for each topic word set, and to use the word feature vectors and named entity feature vectors of each topic word set to perform clustering processing on each topic word set to obtain multiple news topic clusters.

[0122] The hot word identification unit is used to calculate the word popularity of each topic word in each news topic cluster, and based on the word popularity of each topic word in each news topic cluster, to determine the hot words of each news topic cluster.

[0123] The push unit is used to push trending keywords from various news topic clusters to the news dissemination platform to complete the push of trending keywords for news dissemination.

[0124] like Figure 3 As shown, the third aspect of this embodiment provides a news publicity and communication management device. Taking the device as an electronic device as an example, it includes: a memory, a processor, and a transceiver that are connected in sequence. The memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the news publicity and communication management method as described in the first aspect of the embodiment.

[0125] For specific examples, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0126] In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. For example, the processor may not be limited to microprocessors of the STM32F105 series, reduced instruction set computer (RISC) microprocessors, x86 architecture processors, or processors with integrated neural network processing units (NPUs). The transceiver may be, but is not limited to, a Wi-Fi transceiver, a Bluetooth transceiver, a General Packet Radio Service (GPRS) transceiver, a ZigBee (a low-power LAN protocol based on the IEEE 802.15.4 standard) transceiver, a 3G transceiver, a 4G transceiver, and / or a 5G transceiver. Furthermore, the device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0127] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0128] The fourth aspect of this embodiment provides a storage medium that stores instructions containing the news publicity and communication management method described in the first aspect of the embodiment. That is, the storage medium stores instructions that, when executed on a computer, perform the news publicity and communication management method as described in the first aspect of the embodiment.

[0129] The storage medium refers to a carrier for storing data, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0130] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0131] The fifth aspect of this embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the news publicity and communication management method as described in the first aspect of the embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0132] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for managing news and publicity communications, characterized in that, include: Obtain all news text data from various news platforms within a preset time window; Each news text data is processed for keyword identification to obtain the keyword set corresponding to each news text data; Named entity recognition is performed on each topic term set to obtain the named entity feature vector of each topic term set; The word feature vectors of each topic word set are constructed, and the word feature vectors and named entity feature vectors of each topic word set are used to perform clustering processing on each topic word set to obtain multiple news topic clusters; Calculate the word popularity of each topic word in each news topic cluster, and determine the hot words of each news topic cluster based on the word popularity of each topic word in each news topic cluster; Push trending keywords from various news topic clusters to the news dissemination platform to complete the push of trending keywords for news dissemination.

2. The method according to claim 1, characterized in that, Each news text is processed for keyword identification to obtain a set of keywords corresponding to each news text, including: The preset time window is divided into multiple sampling windows, and each news text data is divided into different sampling windows to obtain the news text data in each sampling window; For any news text data within the k-th sampling window, perform word segmentation on any news text data within the k-th sampling window to obtain a word segmentation set, where the initial value of k is 1; Obtain several consecutive sampling windows preceding the k-th sampling window as the target window; Based on the news text data within the target window, the topic weight of each word in the word segmentation set is calculated; Based on the topic weight of each word, several words are selected from the word segmentation set to form the topic word set of any news text data. After all the news text data in the kth sampling window have been queried, the topic word set of each news text data in the kth sampling window is obtained. Increment k by 1 and re-segment any news text data in the k-th sampling window until k equals K, thus obtaining the topic word set corresponding to each news text data, where K is the total number of sampling windows.

3. The method according to claim 2, characterized in that, Based on the news text data within the target window, the topic weight of each word in the word segmentation set is calculated, including: For any word in the word segmentation set, the first frequency of occurrence of the word in all news text data within the k-th sampling window and the second frequency of occurrence in all news text data within each target window are calculated. Based on the frequency of each second occurrence, the attention characteristics of any of the above words are calculated; Based on each second occurrence frequency and the first occurrence frequency, the word frequency change characteristics of any word are calculated; Using the first occurrence frequency, the attention feature, and the word frequency change feature, the topic weight of any word is calculated, and after all words in the word segmentation set have been queried, the topic weight of each word is obtained.

4. The method according to claim 1, characterized in that, By utilizing the word feature vectors and named entity feature vectors of each topic term set, clustering is performed on each topic term set to obtain multiple news topic clusters, including: For the j-th keyword set, obtain the clustering addition time of the j-th keyword set; Obtain the word feature vectors, named entity feature vectors, and most recent update time for each historical theme cluster in the historical cluster set; Based on the clustering addition time and the most recent update time of each historical topic cluster, invalid topic clusters are deleted from the historical cluster set, so that the remaining historical topic clusters are regarded as valid topic clusters; Based on the word feature vector and named entity feature vector of the j-th topic word set, and the word feature vector and named entity feature vector of each effective topic cluster, the insertion cluster of the j-th topic word set is determined from each effective topic cluster; Add the j-th keyword set to the insertion cluster to obtain the initial news topic cluster; The inserted clusters in the effective cluster set are replaced with the initial news topic clusters to obtain a new effective cluster set, wherein the effective cluster set contains all effective topic clusters; The historical cluster set is updated to a new valid cluster set, and j is incremented by 1. The clustering addition time of the j-th topic word set is re-acquired until j equals m, resulting in several initial news topic clusters. The initial value of j is 1, and m is the total number of topic word sets. Several initial news topic clusters are merged to obtain multiple news topic clusters after merging.

5. The method according to claim 4, characterized in that, Based on the word feature vector and named entity feature vector of the j-th topic word set, and the word feature vector and named entity feature vector of each effective topic cluster, the insertion cluster of the j-th topic word set is determined from each effective topic cluster, including: Based on the word feature vector and named entity feature vector of the j-th topic word set, and the word feature vector and named entity feature vector of the ith effective topic cluster, the text similarity and named entity similarity between the j-th topic word set and the ith effective topic cluster are calculated. Determine whether the i-th valid topic cluster satisfies the clustering insertion condition, wherein the clustering insertion condition is that the text similarity is greater than the maximum text similarity, or the named entity similarity is greater than the entity similarity threshold; If not, increment i by 1 and recalculate the text similarity and named entity similarity between the j-th topic word set and the i-th effective topic cluster until the i-th effective topic cluster meets the clustering insertion condition. Then, the i-th effective topic cluster that meets the clustering insertion condition is taken as the inserted cluster.

6. The method according to claim 5, characterized in that, When all valid topic clusters have been polled, and the last valid topic cluster still does not meet the clustering insertion condition, the method further includes: Obtain the text similarity and named entity similarity between the j-th topic term set and each valid topic cluster; Determine whether the similarity of each text is less than the minimum text similarity, and whether the similarity of each named entity is less than the entity similarity threshold; If not, sum the text similarity and named entity similarity between the j-th topic word set and any valid topic cluster to obtain the total similarity. After iterating through all valid topic clusters, several total similarities are obtained. From several total similarity scores, the effective topic cluster corresponding to the highest total similarity score is selected as the insertion cluster.

7. The method according to claim 4, characterized in that, Several initial news topic clusters are merged to obtain multiple news topic clusters after merging, including: For any initial news topic cluster, calculate the word popularity of each topic word in that initial news topic cluster; Based on the word popularity of each topic word in any initial news topic cluster, the total popularity of any initial news topic cluster is determined, and after all initial news topic clusters have been queried, the total popularity of each initial news topic cluster is obtained. From a number of initial news topic clusters, remove those whose total popularity is less than the popularity threshold, and use the remaining initial news topic clusters to form a pre-selected topic cluster set; Merge similar pre-selected topic clusters in the pre-selected topic cluster set to obtain multiple news topic clusters after merging.

8. The method according to claim 7, characterized in that, Calculate the word popularity of each topic term in any initial news topic cluster, including: For any keyword in any initial news topic cluster, count the first occurrence of the keyword in the initial news topic cluster and the second occurrence of the keyword in all news text data; Obtain the average word count of the news text data corresponding to all keyword sets contained in any initial news topic cluster; From all news text data, select news text data containing any of the aforementioned keywords as target news; Obtain the number of reposts, views, and comments for each target news item, and calculate the popularity of each target news item based on these numbers. The word popularity of any given topic is calculated based on the popularity of each target news item, the number of first occurrences, the number of second occurrences, and the average number of words.

9. A news and publicity communication management system, characterized in that, include: The acquisition unit is used to acquire all news text data from various news platforms within a preset time window. The topic word recognition unit is used to perform topic word recognition processing on each news text data to obtain the topic word set corresponding to each news text data; The named entity recognition unit is used to perform named entity recognition on each topic word set to obtain the named entity feature vector of each topic word set; Clustering units are used to construct word feature vectors for each topic word set, and to use the word feature vectors and named entity feature vectors of each topic word set to perform clustering processing on each topic word set to obtain multiple news topic clusters; The hot word recognition unit is used to calculate the word popularity of each topic word in each news topic cluster, and to determine the hot words of each news topic cluster based on the word popularity of each topic word in each news topic cluster. The push unit is used to push trending keywords from various news topic clusters to the news dissemination platform to complete the push of trending keywords for news dissemination.

10. A computer program product containing instructions, characterized in that, When the instructions are executed on the computer, the computer performs the news publicity and communication management method as described in any one of claims 1 to 8.

Images