A topic evolution analysis method, device and equipment

Abstract

The application relates to the technical field of data mining, in particular to a topic evolution analysis method, device and equipment, which are used for solving the problem of inaccurate topic evolution analysis of a traditional topic evolution analysis model, and the method is as follows: a topic evolution analysis model is used to extract each topic from corpus in any historical time window in continuous multiple historical time windows, and to determine the first actual topic intensity of each topic in the historical time window; for any topic, according to the first predicted topic intensity of the topic in a current time window and the second actual topic intensity of the topic in the current time window, a fusion value of a fusion time window is determined, and according to the fusion value, the hyperparameter of the topic in the current time window in the topic evolution analysis model is determined; then, the topic evolution analysis model after each hyperparameter is determined is used to extract topics from corpus obtained in the current time window; in this way, more accurate topic evolution analysis is realized.

Description

Technical Field

[0001] This application relates to the field of data mining technology, and in particular to a method, apparatus and equipment for topic evolution analysis. Background Technology

[0002] Text mining is the process of extracting previously unknown, understandable, and ultimately usable knowledge from large amounts of unstructured text data; it is essentially a database-based data mining and knowledge discovery process. With the rapid development of artificial intelligence, smart homes can use text mining on historical interaction corpora to determine user preferences, thereby providing more personalized and reasonable recommendations. For example, a smart air conditioner can analyze user behavior and preferences by acquiring information such as temperature and fan speed used within a certain time window, allowing for targeted recommendations of suitable temperatures and fan speeds in subsequent time windows. Similarly, a smart refrigerator can analyze user interaction data within a certain time window to determine dietary preferences (e.g., frequently eating vegetables or specific types of vegetables), enabling targeted recommendations of ingredients or recipes that align with those preferences, or nutritional balance tips, in the next time window.

[0003] With existing technology, topic modeling is an important text mining technique. Traditional topic modeling algorithms are often unsupervised machine learning algorithms, such as Latent Dirichlet Allocation (LDA) models and Online Latent Dirichlet Allocation (OLDA) models.

[0004] Those skilled in the art will know that LDA models can analyze the topic situation in a corpus, but cannot incrementally analyze the topic situation of newly arrived texts. The OLDA model that followed can obtain the prior parameters of the current time window by constructing the topic content evolution matrix of past time windows, so as to analyze the evolution of topic content and intensity online. However, OLDA models have problems such as mixing of old and new topics, online analysis effect depending on weight setting, and slow convergence speed, which leads to inaccurate topic evolution analysis. Summary of the Invention

[0005] This application provides a method, apparatus, and device for topic evolution analysis to improve the accuracy of topic evolution analysis.

[0006] The specific technical solutions provided in this application are as follows:

[0007] In a first aspect, embodiments of this application provide a topic evolution analysis method, including:

[0008] For any historical time window among multiple consecutive historical time windows, a topic evolution analysis model is used to extract each topic in the corpus within the historical time window, and the first actual topic intensity of each topic in the historical time window is determined, wherein the size of the historical time window is obtained in advance by adjusting the window size using the topic evolution analysis model;

[0009] For any given topic, based on the first actual topic intensity of the topic within each historical time window, the first predicted topic intensity of the topic within the current time window is determined. Based on each first predicted topic intensity and the second actual topic intensity of the topic within the current time window, the fusion value of the fusion time window corresponding to the topic is determined. Based on the fusion value, the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window are determined.

[0010] The topic evolution analysis model, with each hyperparameter determined, is used to extract topics from the corpus acquired within the current time window.

[0011] In one possible implementation, determining the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window based on the fusion value includes:

[0012] Based on the fused values, determine the prior parameter weight matrix corresponding to the current time window;

[0013] Based on the prior parameter weight matrix and the topic content evolution matrix corresponding to the current time window, determine the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window.

[0014] In one possible implementation, determining the fusion value of the fusion time window corresponding to the topic based on each first predicted topic intensity and the second actual topic intensity of the topic within the current time window includes:

[0015] Based on the intensity of each first predicted topic and the intensity of the second actual topic within the current time window, the target deviation value corresponding to the topic is determined;

[0016] The fusion value for the required fusion time window corresponding to the topic is determined based on the target deviation value.

[0017] In one possible implementation, determining the fusion value of the fusion time window corresponding to the topic based on the target deviation value includes:

[0018] If the target deviation value is not greater than the first threshold, then the sum of the fusion value corresponding to the previous historical time window of the current time window and the preset value is determined as the fusion value of the time window to be fused corresponding to the topic.

[0019] If the target deviation value is not less than the second threshold, then the difference between the fusion value corresponding to the previous historical time window of the current time window and the preset value is determined as the fusion value of the time window to be fused corresponding to the topic.

[0020] If the target deviation value is greater than the first threshold and less than the second threshold, then the fusion value corresponding to the previous historical time window of the current time window is determined as the fusion value of the time window to be fused corresponding to the topic.

[0021] In one possible implementation, determining the target deviation value corresponding to the topic based on each first predicted topic intensity and the second actual topic intensity of the topic within the current time window includes:

[0022] Determine the absolute value of the difference between the intensity of each first predicted topic and the intensity of the second actual topic within the current time window;

[0023] From all the absolute values, select the preset number of absolute values ​​that are ranked first in descending order;

[0024] The target deviation value corresponding to the topic is determined based on the absolute value of the preset quantity.

[0025] In one possible implementation, the method further includes:

[0026] The document-topic distribution for multiple consecutive target time windows is obtained using the aforementioned topic evolution analysis model, and a line graph of topic intensity corresponding to the multiple target time windows is plotted; and / or

[0027] The topic evolution analysis model is used to obtain the topic-word distribution of multiple consecutive target time windows, and the topic content evolution diagrams corresponding to the multiple target time windows are drawn.

[0028] Display the theme intensity line graph and / or the theme content evolution graph in the designated display interface.

[0029] In one possible implementation, the size of the historical time window is obtained as follows:

[0030] For any initial time window among multiple consecutive initial time windows, the topic evolution analysis model is used to extract each topic in the corpus within the initial time window, and the third actual topic intensity of each topic in the initial time window is determined. The corpus within the initial time window is obtained by dividing the corpus within a preset time period based on window division rules.

[0031] For any given topic, based on the third actual topic intensity within each initial time window, a second predicted topic intensity within the historical time window is determined, and based on the second predicted topic intensity and the fourth actual topic intensity within the historical time window, the deviation of the topic is determined; wherein, the historical time window is adjacent to and follows the plurality of initial time windows;

[0032] The size of the historical time window is adjusted according to the deviation of each topic.

[0033] In one possible implementation, the topic evolution analysis model is an adaptive online topic evolution DAOLDA model, which is obtained by improving the online implicit Dirichlet distribution OLDA model.

[0034] Secondly, embodiments of this application provide a topic evolution analysis apparatus, comprising:

[0035] The first determining module is used to extract each topic in the corpus within any historical time window from multiple consecutive historical time windows using a topic evolution analysis model, and to determine the first actual topic intensity of each topic in the historical time window, wherein the size of the historical time window is obtained in advance by adjusting the window size using the topic evolution analysis model.

[0036] The second determining module is used to determine, for any topic, the first predicted topic intensity of the topic in the current time window based on the first actual topic intensity of the topic in each historical time window, and to determine the fusion value of the time window to be fused corresponding to the topic based on each first predicted topic intensity and the second actual topic intensity of the topic in the current time window, and to determine the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window based on the fusion value.

[0037] The topic extraction module is used to extract topics from the corpus acquired within the current time window using a topic evolution analysis model after each hyperparameter is determined.

[0038] In one possible implementation, the second determining module is used to determine the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window based on the fusion value.

[0039] Based on the fused values, determine the prior parameter weight matrix corresponding to the current time window;

[0040] Based on the prior parameter weight matrix and the topic content evolution matrix corresponding to the current time window, determine the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window.

[0041] In one possible implementation, the second determining module is used to determine the fusion value of the time window to be fused corresponding to the topic based on each first predicted topic intensity and the second actual topic intensity of the topic within the current time window.

[0042] Based on the intensity of each first predicted topic and the intensity of the second actual topic within the current time window, the target deviation value corresponding to the topic is determined;

[0043] The fusion value for the required fusion time window corresponding to the topic is determined based on the target deviation value.

[0044] In one possible implementation, the second determining module is used to: determine the fusion value of the fusion time window corresponding to the topic based on the target deviation value.

[0045] If the target deviation value is not greater than the first threshold, then the sum of the fusion value corresponding to the previous historical time window of the current time window and the preset value is determined as the fusion value of the time window to be fused corresponding to the topic.

[0046] If the target deviation value is not less than the second threshold, then the difference between the fusion value corresponding to the previous historical time window of the current time window and the preset value is determined as the fusion value of the time window to be fused corresponding to the topic.

[0047] If the target deviation value is greater than the first threshold and less than the second threshold, then the fusion value corresponding to the previous historical time window of the current time window is determined as the fusion value of the time window to be fused corresponding to the topic.

[0048] In one possible implementation, the second determining module is used to determine the target deviation value corresponding to the topic based on each first predicted topic intensity and the second actual topic intensity of the topic within the current time window, wherein the second determining module is configured to:

[0049] Determine the absolute value of the difference between the intensity of each first predicted topic and the intensity of the second actual topic within the current time window;

[0050] From all the absolute values, select the preset number of absolute values ​​that are ranked first in descending order;

[0051] The target deviation value corresponding to the topic is determined based on the absolute value of the preset quantity.

[0052] In one possible implementation, the topic extraction module is further configured to:

[0053] The document-topic distribution for multiple consecutive target time windows is obtained using the aforementioned topic evolution analysis model, and a line graph of topic intensity corresponding to the multiple target time windows is plotted; and / or

[0054] The topic evolution analysis model is used to obtain the topic-word distribution of multiple consecutive target time windows, and the topic content evolution diagrams corresponding to the multiple target time windows are drawn.

[0055] Display the theme intensity line graph and / or the theme content evolution graph in the designated display interface.

[0056] In one possible implementation, the size of the historical time window is obtained as follows:

[0057] For any initial time window among multiple consecutive initial time windows, the topic evolution analysis model is used to extract each topic in the corpus within the initial time window, and the third actual topic intensity of each topic in the initial time window is determined. The corpus within the initial time window is obtained by dividing the corpus within a preset time period based on window division rules.

[0058] For any given topic, based on the third actual topic intensity within each initial time window, a second predicted topic intensity within the historical time window is determined, and based on the second predicted topic intensity and the fourth actual topic intensity within the historical time window, the deviation of the topic is determined; wherein, the historical time window is adjacent to and follows the plurality of initial time windows;

[0059] The size of the historical time window is adjusted according to the deviation of each topic.

[0060] In one possible implementation, the topic evolution analysis model is an adaptive online topic evolution DAOLDA model, which is obtained by improving the online implicit Dirichlet distribution OLDA model.

[0061] Thirdly, embodiments of this application provide an electronic device, including:

[0062] Memory is used to store computer programs that can be executed by the controller;

[0063] The controller is connected to the memory and is configured to perform the method as described in any of the first aspects above.

[0064] Fourthly, embodiments of this application provide a computer-readable storage medium that, when instructions in the storage medium are executed by a processor, enables the processor to perform the method described in any one of the first aspects above.

[0065] Fifthly, this application provides a computer program product comprising: computer program code, which, when run on a computer, causes the computer to perform the method described in any one of the first aspects.

[0066] The technical solution provided in this application provides at least the following beneficial effects: Based on the first predicted topic strength and the second actual topic strength of any topic within the current time window, the fusion value of the time window to be fused corresponding to the topic is determined. Based on the fusion value, the prior parameter weights in the topic content evolution matrix corresponding to the current time window are updated, realizing topic alignment in different time windows, ensuring the consistency of topics in different time windows, and achieving more accurate topic evolution analysis. Furthermore, the above-mentioned topic evolution analysis model utilizes a variable sliding time window mechanism, which can continuously update the hyperparameters of the topic-word distribution in the model to integrate previously learned knowledge into the topic evolution analysis model corresponding to the new time window. Thus, without retaining historical data, some previously acquired prior knowledge is retained, realizing online learning, knowledge fusion, and adaptive adjustment of topic evolution in the topic evolution analysis model, thereby achieving more accurate topic evolution analysis. Attached Figure Description

[0067] Figure 1 This is a schematic diagram of a probabilistic graphical representation of an OLDA model in an embodiment of this application;

[0068] Figure 2 This is a schematic diagram of a probability graph representation of an AOLDA model in an embodiment of this application;

[0069] Figure 3 This is a schematic diagram of a probability graph representation of a DAOLDA model in an embodiment of this application;

[0070] Figure 4This is a flowchart illustrating a topic evolution analysis method in an embodiment of this application;

[0071] Figure 5 This is a flowchart illustrating a method for adaptively adjusting a time window according to an embodiment of this application.

[0072] Figure 6 This is a schematic diagram of a process for determining the fusion value of the fusion time window in an embodiment of this application;

[0073] Figure 7 This is a schematic diagram of a process for determining the target deviation value corresponding to a subject in an embodiment of this application;

[0074] Figure 8 This is a flowchart illustrating a method for determining hyperparameters corresponding to a topic in an embodiment of this application.

[0075] Figure 9 This is a flowchart illustrating a method for displaying the results of a topic evolution analysis in an embodiment of this application;

[0076] Figure 10 This is a schematic diagram of a theme intensity line graph in an embodiment of this application;

[0077] Figure 11 This is a schematic diagram of the logical architecture of a topic evolution analysis device in an embodiment of this application;

[0078] Figure 12 This is a schematic diagram of the physical architecture of the electronic device in the embodiments of this application. Detailed Implementation

[0079] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0080] It should be noted that the terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in sequences other than those illustrated or described herein.

[0081] The design concept of the embodiments of this application will be briefly introduced below.

[0082] This application relates to the field of data mining technology, and mainly to a method, apparatus and equipment for topic evolution analysis.

[0083] The LDA model, proposed by Blei et al., is a three-layer Bayesian model, an unsupervised probabilistic generative model capable of representing text topics. The three layers refer to the text, topics, and vocabulary. This model assumes that text is a mixture of several implicit topics, and topics are a mixture of all words in the vocabulary. Different texts differ in their topic mixing ratios and are sampled from a Dirichlet distribution.

[0084] The generation process of an LDA model is a sampling process that describes how words in a document are generated under hidden topics. In LDA, a topic is represented by a set of semantically related words and the probability of those words appearing on that topic. Typically, based on a given corpus, the document-topic distribution θ and the topic-word distribution are trained using Gibbs sampling. This forms the basis for topic evolution analysis. However, the LDA model cannot incrementally analyze the status of newly arriving text topics.

[0085] Figure 1 A schematic diagram of the probabilistic graphical representation of the OLDA model is shown. (See also...) Figure 1 As shown, the OLDA model is an improved version of LDA, which is based on online text analysis. It introduces the concept of time discretization into LDA, dividing the text into continuous time windows for modeling. The most important assumption of the OLDA model is topic inheritance, meaning that topics have heritability and stability, and the text topic information in historical time windows can influence subsequent time windows, mainly manifested in the continued existence of topics across multiple time windows.

[0086] Based on this, the OLDA model uses the topic-word distribution within historical time windows to construct a topic content evolution matrix, storing historical topic information in this matrix and assigning weight vectors to it to measure the influence of different historical time windows on the topic of the current time window, thereby obtaining the prior parameters of the current time window. Maintain the continuity of the theme. Prior parameters. The construction formula is as follows:

[0087]

[0088] in, Let be the prior parameters for the k-th topic within the t-th time window. Let ω be the topic content evolution matrix of topic k over δ time windows. δ This represents the weight vector for different time windows.

[0089] However, the aforementioned OLDA model suffers from problems such as mixing old and new topics, online analysis performance depending on weight settings, and slow convergence speed, leading to inaccurate topic evolution analysis. Furthermore, traditional OLDA models typically use static time windows, i.e., time windows with fixed sizes. This often results in some time windows acquiring more data than others. When performing topic evolution analysis on documents based on these static time windows, document overfitting frequently occurs, further exacerbating the inaccuracy of topic evolution analysis.

[0090] In view of this, in order to solve the problem of inaccurate topic evolution analysis in traditional topic evolution analysis models, in this embodiment, for any historical time window in multiple consecutive historical time windows, a topic evolution analysis model is used to extract each topic in the corpus within that historical time window, and the first actual topic intensity of each topic in the historical time window is determined. The size of the historical time window is obtained by pre-adjusting the window size using the topic evolution analysis model. Then, for any topic, based on the first predicted topic intensity of the topic in the current time window determined by the first actual topic intensity of the topic in each historical time window, and the second actual topic intensity of the topic in the current time window, the fusion value of the fusion time window corresponding to the topic is determined, and based on the fusion value, the hyperparameters in the topic evolution analysis model corresponding to the topic in the current time window are determined. Finally, the topic evolution analysis model with each hyperparameter determined is used to extract topics from the corpus obtained in the current time window.

[0091] The topic evolution analysis method provided in this application can determine the fusion value of the corresponding fusion time window based on the first predicted topic strength and the second actual topic strength of any topic in the current time window. Based on the fusion value, the prior parameter weights in the topic content evolution matrix corresponding to the current time window are updated, realizing topic alignment in different time windows, ensuring the consistency of topics in different time windows, and achieving more accurate topic evolution analysis.

[0092] Furthermore, the aforementioned topic evolution analysis model utilizes a variable sliding time window mechanism, which can continuously update the hyperparameters of the topic-word distribution in the model to integrate previously learned knowledge into the topic evolution analysis model corresponding to the new time window. This allows for the retention of some previously acquired prior knowledge without retaining historical data, enabling online learning, knowledge fusion, and adaptive adjustment of the topic evolution analysis model, thereby achieving more accurate topic evolution analysis.

[0093] In this embodiment, the topic evolution analysis model is an improvement upon the OLDA model. To address the overfitting problem of documents within the static time window, the inventors improved the OLDA model to obtain an Adaptive Online Latent Dirichlet Allocation (AOLDA) model, enabling adaptive adjustment of the time window size. Figure 2 , Figure 2 A schematic diagram of the probability graph representation of the AOLDA model described above is shown.

[0094] To further improve the accuracy of topic evolution analysis in the topic evolution analysis model and to better and more rationally integrate previously learned knowledge into the topic evolution analysis model corresponding to new time windows, an adaptive online topic evolution model (Dynamic Adaptive Online Latent Dirichlet Allocation, DAOLDA) is further improved based on the AOLDA model. This model utilizes a variable sliding time window mechanism to continuously update the hyperparameters of the topic-word distribution in the model, integrating previously learned knowledge into the topic evolution analysis model corresponding to new time windows, thus achieving adaptive adjustment to topic evolution. Figure 3 , Figure 3 A schematic diagram of the probability graph representation of the DAOLDA model is shown.

[0095] After introducing the design concept of the embodiments of this application, the preferred embodiments of this application will be further described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for illustration and explanation of this application and are not intended to limit this application. Furthermore, the embodiments of this application and the features in the embodiments can be combined with each other without conflict.

[0096] See Figure 4 As shown in the embodiments of this application, a topic evolution analysis method is provided. The specific process of this method is as follows:

[0097] Step 400: For any historical time window among multiple consecutive historical time windows, use the topic evolution analysis model to extract each topic in the corpus within the historical time window, and determine the first actual topic intensity of each topic in the historical time window. The size of the historical time window is obtained by adjusting the window size in advance using the topic evolution analysis model.

[0098] See also Figure 2 As shown in this embodiment, the hyperparameter update of topic k in the AOLDA model is as follows:

[0099] Assuming a time window t=1, what are the parameters of the topic-word distribution in the topic evolution analysis model? It is obtained from Dirichlet prior.

[0100] So, Wherein, the weight vector β 1,k Each element value is initialized to b.

[0101] Because the AOLDA model is used to adaptively divide the time window t, the topic evolution analysis model will handle the variable length M. t Document flow of size Among them, S t Document Stream and Document Sequence in the t-th Time Window Maintain consistency.

[0102] Therefore, the document d obtained within the time window t can be represented by word class vectors. S indicates. t The document stream adds a new set of words to the vocabulary, assuming that these new words were present in the previous document stream S. t-1 The word distribution count is 0 for all topics.

[0103] The prior parameters in the aforementioned OLDA model The construction formula is the same. In the AOLDA model, the hyperparameters corresponding to topic k in time window t are derived from a weighted mixture of past topics:

[0104]

[0105] In this embodiment of the application, after describing the hyperparameter update of the AOLDA model above, before executing step 200, please refer to... Figure 5 As shown, the following steps are also required to achieve adaptive adjustment of the time window:

[0106] Step 500: For any initial time window among multiple consecutive initial time windows, use the topic evolution analysis model to extract each topic in the corpus within the initial time window, and determine the third actual topic strength of each topic in the initial time window. The corpus within the initial time window is obtained by dividing the corpus within a preset time period based on the window division rules.

[0107] In practice, before executing step 500, the following operations also need to be performed:

[0108] Operation 1: Obtain the corpus within a preset time period. This corpus can be text data generated through various methods such as image interaction, voice interaction, and large-screen interaction.

[0109] In this embodiment of the application, after obtaining the corpus within a continuous preset time period, it is also necessary to sort them in ascending order according to their publication time.

[0110] The second step involves preprocessing the acquired corpus by removing punctuation, stop words, and word segmentation, thereby completing the cleaning and statistical analysis of the corpus.

[0111] Operation 3: Based on the window partitioning rules, the preprocessed corpus is partitioned into windows to obtain multiple discrete initial time windows.

[0112] In this embodiment of the application, the above window division rules are not specifically limited. They can be customized according to the timeliness and quantity of the corpus and according to actual needs, such as the initial time window size being a fixed value.

[0113] Step 510: For any topic, determine the second predicted topic strength of the topic in the historical time window based on the third actual topic strength of the topic in each initial time window, and determine the deviation of the topic based on the second predicted topic strength and the fourth actual topic strength of the topic in the historical time window; wherein, the historical time window is adjacent to and follows multiple initial time windows.

[0114] Step 520: Adjust the size of the historical time window according to the deviation of each topic to obtain the size of the historical time window.

[0115] In this way, based on the deviation between the intensity of the second predicted theme and the intensity of the fourth actual theme, the reasonableness of the size of the historical time window is determined. If the deviation value determined based on the deviation status of all themes extracted from any initial time window is not less than the first preset threshold, the size of the historical time window is reduced; if the deviation value determined based on the deviation status of all themes extracted from any initial time window is less than or equal to the second preset threshold, the size of the historical time window is increased. In this way, the size of the historical time window can be adaptively adjusted, thereby improving the flexibility of adjusting the size of the time window.

[0116] It should be noted that the above deviation values ​​measure the difference between the predicted theme strength and the actual theme strength within a historical time window, that is, the degree of fluctuation in the actual theme strength relative to the predicted theme strength. A larger deviation value indicates greater fluctuation in the historical time window, in which case the size of the historical time window should be reduced; a smaller deviation value indicates less fluctuation in the historical time window, in which case the size of the historical time window should be increased.

[0117] It should also be noted that the above deviation values ​​were determined by selecting a certain number of deviations from the predicted topic strength and the fourth actual topic strength within the historical time window, in descending order, to determine the corresponding fourth actual topic strengths. By selecting the second predicted topic strength and the fourth actual topic strength within the historical time window for each topic with larger fluctuations to adjust the historical time window, the size of the historical time window can be made more reasonable and accurate, thereby ensuring that the fusion values ​​for the fusion time window determined based on the corpus within the historical time window are also more accurate.

[0118] Step 410: For any topic, determine the first predicted topic strength in the current time window based on the first actual topic strength in each historical time window.

[0119] For example, suppose we use a topic evolution analysis model to extract three topics from any historical time window, which are denoted as topic 1, topic 2 and topic 3 respectively. These topics 1, topic 2 and topic 3 can be pre-defined.

[0120] Therefore, for any of the above topics, topic extraction is performed based on the corpus obtained within each historical time window.

[0121] Let's assume the historical time windows are historical time window 1, historical time window 2, and historical time window 3, and that a1 words are extracted from historical time window 1 for topic 1, a2 words are extracted from historical time window 2 for topic 1, and a3 words are extracted from historical time window 3 for topic 1.

[0122] Therefore, the first actual topic intensity of topic 1 in each historical time window is determined. For example, for historical time window 1, the first actual topic intensity of topic 1 in historical time window 1 can be represented by the ratio of the number of words corresponding to topic 1 to the sum of the number of words of all topics. Alternatively, other weight values ​​can be introduced to calculate the first actual topic intensity. In this embodiment, the method for determining the first actual topic intensity is not specifically limited.

[0123] In this embodiment of the application, after obtaining the first actual topic intensity of any topic in each historical time window, the topic intensity of the topic in the current time window is predicted based on the multiple first actual topic intensities, thereby obtaining the first predicted topic intensity of the topic in the current time window.

[0124] In this embodiment, the least squares method can be used to fit the intensity of each first actual topic, and the first predicted topic intensity of the topic in the current time window can be predicted based on the fitting result; alternatively, a time series model can be used to calculate the first predicted topic intensity of the topic in the current time window; or a time series model can be used to make predictions, and a more reasonable and accurate predicted topic intensity can be obtained based on the evolution trend of topic intensity.

[0125] It should be noted that, in the embodiments of this application, the above prediction method is not specifically limited, and other methods can also be used to obtain the first predicted topic intensity.

[0126] Step 420: Based on the intensity of each first predicted topic and the intensity of the second actual topic within the current time window, determine the fusion value of the time window to be fused for each topic.

[0127] In this embodiment of the application, when performing step 420, refer to... Figure 6 As shown, the specific steps include the following:

[0128] Step 4201: Determine the target deviation value corresponding to each topic based on the intensity of each first predicted topic and the intensity of the second actual topic within the current time window.

[0129] In the embodiments of this application, see the following: Figure 7 As shown, when executing step 4201, the following steps are specifically performed:

[0130] Step 700: Determine the absolute value of the difference between the intensity of each first predicted topic and the intensity of the second actual topic within the current time window.

[0131] Step 710: Select the preset number of absolute values ​​that are ranked first from the absolute values ​​in descending order.

[0132] In this embodiment, during step 700, the absolute value of the difference between the intensity of any first predicted topic and the intensity of the second actual topic is calculated. During step 710, a preset number of absolute values ​​are selected from the calculated absolute values, which are sorted in descending order. By selecting the absolute values ​​with larger fluctuations in any topic, topic alignment can be achieved across different time windows, ensuring the continuity and consistency of topics across different time windows. This enables online learning of the topic evolution analysis model, improves the accuracy of topic extraction in continuous time, and makes the recommendations generated based on the topic extraction results more accurate and user-friendly.

[0133] Step 720: Determine the target deviation value corresponding to the topic based on the absolute value of the preset quantity.

[0134] In this embodiment of the application, when performing step 720, the target deviation value corresponding to the above-mentioned topic can be obtained by taking the average, median, or other methods of the absolute values ​​of the preset quantity, and no restrictions are imposed here.

[0135] Step 4202: Based on the target deviation value, determine the fusion value of the fusion time window corresponding to the theme.

[0136] In this embodiment of the application, when performing step 4202, the following three cases are included, but not limited to, depending on the magnitude of the target deviation value:

[0137] In case one, if the target deviation value is not greater than the first threshold, the sum of the fused value corresponding to the previous historical time window of the current time window and the preset value is determined as the fused value of the time window to be fused corresponding to the theme.

[0138] In the second scenario, if the target deviation value is not less than the second threshold, the difference between the fusion value corresponding to the previous historical time window of the current time window and the preset value is determined as the fusion value of the time window to be fused corresponding to the theme.

[0139] Scenario 3: If the target deviation value is greater than the first threshold and less than the second threshold, then the fusion value corresponding to the previous historical time window of the current time window is determined as the fusion value of the time window to be fused corresponding to the theme.

[0140] In this embodiment of the application, the above three cases can be represented by the following formula:

[0141]

[0142] Where, N t+1 N represents the fusion value corresponding to the time window (t+1). t E represents the fusion value corresponding to time window t. t η1 is the target deviation value between the first predicted topic intensity and the second actual topic intensity within a time window (t+1) for any topic. η1 is the first threshold and η2 is the second threshold. The preset value can be 1, and η1 and η2 can be set according to actual needs.

[0143] Thus, using N t+1 A historical time window (i.e., (tN) t+1 The statistical data in (t+1, ..., t) are updated to the prior parameter values ​​of the topic-word distribution within the time window (t+1), and the hyperparameter β in the topic evolution analysis model corresponding to the time window (t+1) is updated. t+1 The value of is used to integrate the previously learned knowledge of the topic evolution analysis model into the topic evolution analysis model corresponding to the new time window, that is, the topic evolution analysis model corresponding to the current time window mentioned above.

[0144] Step 430: Based on the fusion values, determine the hyperparameters in the topic evolution analysis model corresponding to the topic in the current time window.

[0145] In the embodiments of this application, see the following: Figure 8 As shown, during step 430, the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window are determined by performing the following steps:

[0146] Step 4301: Determine the prior parameter weight matrix corresponding to the current time window based on the fused values.

[0147] Step 4302: Determine the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window based on the prior parameter weight matrix and the topic content evolution matrix corresponding to the current time window.

[0148] See also Figure 3 As shown, the hyperparameter β in the theme evolution analysis model corresponding to theme k in time window (t+1) t+1,k It can be expressed by the following formula:

[0149]

[0150] in, N represents t The weight vector of each historical time window has dimension N. t , Let be the topic content evolution matrix corresponding to topic k in the time window (t+1).

[0151] Assume the retrieved documents are sorted in ascending order according to their publication time.

[0152] Therefore, the continuous time corresponding to the above documents is divided into discrete time windows, which are denoted as: 1, ..., t-1, t, where the size of each time window is determined by adaptive adjustment through the AOLDA model mentioned above.

[0153] If time window t contains M t If there are 10 documents, then they can be arranged in chronological order to form a document flow D. t It can be represented as: Of these, d1 is the earliest published document. It is the latest document to be released.

[0154] Assumption For a V t ×N t Matrix, where V t The matrix represents the number of words in the vocabulary corresponding to time window t. The element in the nth row and jth column This indicates that the nth word in the vocabulary is in the (tN)th position. t+1 +j) Count on topic k within the time window.

[0155] And assume N t The value of is 3, and The value is

[0156] So, because N t =3, then β t+1,k The value is determined by the topic-word distribution of the three time windows (t, t-1, t-2).

[0157] because This indicates that the weights are set to be the same and satisfy the constraint that the sum of the weight values ​​is 1.

[0158] Then, β t+1,k It can be expressed by the following formula:

[0159]

[0160] Using the same method described in steps 410-430 above, the hyperparameters in the topic evolution analysis model corresponding to the current time window for each topic can be determined, and after determining the hyperparameters for each topic, step 440 is executed.

[0161] Step 440: Using the topic evolution analysis model after each hyperparameter is determined, extract topics from the corpus acquired within the current time window.

[0162] In some feasible embodiments, as described above, if a new word appears in the document stream that has not appeared before, the new word will be added to the end of the vocabulary, and the prior parameter value of the corresponding topic-word distribution will be initialized to 0, that is, it is assumed that the word distribution count of these new words in all topics of the previous document stream is 0.

[0163] In the embodiments of this application, see the following: Figure 9 As shown, after performing step 440, the following steps can also be performed:

[0164] Step 900: Use the topic evolution analysis model to obtain the document-topic distribution for multiple consecutive target time windows, and draw the topic intensity line graphs corresponding to the multiple target time windows; and / or, use the topic evolution analysis model to obtain the topic-word distribution for multiple consecutive target time windows, and draw the topic content evolution graphs corresponding to the multiple target time windows.

[0165] Step 910: Display a line graph of topic intensity and / or a graph of topic content evolution on the designated display interface.

[0166] For example, see Figure 10 As shown, after extracting topics from multiple target time windows using a topic evolution analysis model, document-topic distributions can be obtained. Then, based on the document-topic distributions, topic intensity line graphs corresponding to multiple target time windows can be plotted. The target time window can be any time window whose size is determined using a variable sliding time window mechanism. Finally, the topic intensity line graphs are displayed on a designated display interface, which allows users to intuitively understand the topic evolution process and their own preferences, thus enabling them to make better adjustments.

[0167] Based on the same inventive concept, see [reference] Figure 11 As shown, this application provides a topic evolution analysis device, including:

[0168] The first determining module 1110 is used to extract each topic in the corpus within any historical time window from a series of consecutive historical time windows using a topic evolution analysis model, and to determine the first actual topic intensity of each topic in the historical time window, wherein the size of the historical time window is obtained by adjusting the window size in advance using the topic evolution analysis model.

[0169] The second determining module 1120 is used to determine, for any topic, a first predicted topic intensity of the topic in the current time window based on the first actual topic intensity of the topic in each historical time window, and to determine the fusion value of the time window to be fused corresponding to the topic based on each first predicted topic intensity and the second actual topic intensity of the topic in the current time window, and to determine the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window based on the fusion value.

[0170] The topic extraction module 1130 is used to extract topics from the corpus acquired within the current time window using a topic evolution analysis model after each hyperparameter is determined.

[0171] In one possible implementation, the second determining module 1120 is used to determine the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window based on the fusion value, wherein the hyperparameters of the topic are determined according to the fusion value.

[0172] Based on the fused values, determine the prior parameter weight matrix corresponding to the current time window;

[0173] Based on the prior parameter weight matrix and the topic content evolution matrix corresponding to the current time window, determine the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window.

[0174] In one possible implementation, the second determining module 1120 is used to determine the fusion value of the fusion time window corresponding to the topic based on each first predicted topic intensity and the second actual topic intensity of the topic within the current time window, wherein the second determining module is configured to:

[0175] Based on the intensity of each first predicted topic and the intensity of the second actual topic within the current time window, the target deviation value corresponding to the topic is determined;

[0176] The fusion value for the required fusion time window corresponding to the topic is determined based on the target deviation value.

[0177] In one possible implementation, the second determining module 1120 is used to: determine the fusion value of the fusion time window corresponding to the topic based on the target deviation value;

[0178] If the target deviation value is not greater than the first threshold, then the sum of the fusion value corresponding to the previous historical time window of the current time window and the preset value is determined as the fusion value of the time window to be fused corresponding to the topic.

[0179] If the target deviation value is not less than the second threshold, then the difference between the fusion value corresponding to the previous historical time window of the current time window and the preset value is determined as the fusion value of the time window to be fused corresponding to the topic.

[0180] If the target deviation value is greater than the first threshold and less than the second threshold, then the fusion value corresponding to the previous historical time window of the current time window is determined as the fusion value of the time window to be fused corresponding to the topic.

[0181] In one possible implementation, the second determining module 1120 is used to determine the target deviation value corresponding to the topic based on each first predicted topic intensity and the second actual topic intensity of the topic within the current time window, wherein the second determining module 1120 is configured to:

[0182] Determine the absolute value of the difference between the intensity of each first predicted topic and the intensity of the second actual topic within the current time window;

[0183] From all the absolute values, select the preset number of absolute values ​​that are ranked first in descending order;

[0184] The target deviation value corresponding to the topic is determined based on the absolute value of the preset quantity.

[0185] In one possible implementation, the topic extraction module 1130 is further configured to:

[0186] The document-topic distribution for multiple consecutive target time windows is obtained using the aforementioned topic evolution analysis model, and a line graph of topic intensity corresponding to the multiple target time windows is plotted; and / or

[0187] The topic evolution analysis model is used to obtain the topic-word distribution of multiple consecutive target time windows, and the topic content evolution diagrams corresponding to the multiple target time windows are drawn.

[0188] Display the theme intensity line graph and / or the theme content evolution graph in the designated display interface.

[0189] In one possible implementation, the size of the historical time window is obtained as follows:

[0190] For any initial time window among multiple consecutive initial time windows, the topic evolution analysis model is used to extract each topic in the corpus within the initial time window, and the third actual topic intensity of each topic in the initial time window is determined. The corpus within the initial time window is obtained by dividing the corpus within a preset time period based on window division rules.

[0191] For any given topic, based on the third actual topic intensity within each initial time window, a second predicted topic intensity within the historical time window is determined, and based on the second predicted topic intensity and the fourth actual topic intensity within the historical time window, the deviation of the topic is determined; wherein, the historical time window is adjacent to and follows the plurality of initial time windows;

[0192] The size of the historical time window is adjusted according to the deviation of each topic.

[0193] In one possible implementation, the topic evolution analysis model is an adaptive online topic evolution DAOLDA model, which is obtained by improving the online implicit Dirichlet distribution OLDA model.

[0194] Based on the same inventive concept, this application provides an electronic device, the structure of which can be as follows: Figure 12As shown, it includes components such as: communication component 1210, memory 1220, display unit 1230, camera 1240, sensor 1250, audio circuit 1260, Bluetooth module 1270, processor 1280, etc.

[0195] The communication component 1210 is used to communicate with a server or other devices. In some embodiments, it may include a Circuit-Based Wireless Fidelity (WiFi) module, which is a short-range wireless transmission technology. Electronic devices can use the WiFi module to help users send and receive information.

[0196] The memory 1220 can be used to store software programs and data. The processor 1280 executes various functions of the electronic device and data processing by running the software programs or data stored in the memory 1220. The memory 1220 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. The memory 1220 stores an operating system that enables the electronic device to run. In this application, the memory 1220 may store the operating system and various application programs, and may also store code that executes the methods of the embodiments of this application.

[0197] The display unit 1230 can also be used to display information input by the user or information provided to the user, as well as a graphical user interface (GUI) for various menus of the electronic device. Specifically, the display unit 1230 may include a display screen 1232 disposed on the front of the electronic device. The display screen 1232 may be configured as a liquid crystal display, a light-emitting diode, or the like. The display unit 1230 can be used to display images and information, etc., as described in the embodiments of this application.

[0198] The display unit 1230 can also be used to receive input digital or character information and generate signal inputs related to user settings and function control of the electronic device. Specifically, the display unit 1230 may include a touch screen 1231 disposed on the front of the electronic device, which can collect touch operations of the user on or near it, such as clicking a button, dragging a scroll box, etc.

[0199] The touchscreen 1231 can be placed on top of the display screen 1232, or the touchscreen 1231 and the display screen 1232 can be integrated to realize the input and output functions of the electronic device. After integration, it can be referred to as a touch display screen. In this application, the display unit 1230 can display the application program and the corresponding operation steps.

[0200] Camera 1240 can be used to capture still or moving images, and users can transmit these images to other devices. There can be one or multiple cameras 1240. An object is projected onto a photosensitive element through a lens, generating an optical image. The photosensitive element can be a charge-coupled device (CCD) or a complementary metal-oxide-semiconductor (CMOS) phototransistor. The photosensitive element converts the light signal into an electrical signal, which is then transmitted to the processor 1280 to be converted into a digital image signal.

[0201] The electronic device may also include at least one sensor 1250, such as an accelerometer 1251, a proximity sensor 1252, a fingerprint sensor 1253, and a temperature sensor 1254. The electronic device may also be equipped with other sensors such as a gyroscope, barometer, hygrometer, thermometer, infrared sensor, light sensor, and motion sensor.

[0202] Audio circuitry 1260, speaker 1261, and microphone 1262 provide an audio interface between the user and the electronic device. Audio circuitry 1260 converts received audio data into electrical signals and transmits them to speaker 1261, where speaker 1261 converts them into sound signals for output. The electronic device can also be equipped with volume buttons for adjusting the volume of the sound signal. On the other hand, microphone 1262 converts collected sound signals into electrical signals, which are then received by audio circuitry 1260, converted into audio data, and output to communication component 1210 for transmission to, for example, another electronic device, or to memory 1220 for further processing.

[0203] The Bluetooth module 1270 is used to interact with other Bluetooth devices that also have a Bluetooth module via the Bluetooth protocol. For example, an electronic device can establish a Bluetooth connection with a wearable electronic device (such as a smartwatch) that also has a Bluetooth module through the Bluetooth module 1270, thereby exchanging data.

[0204] The processor 1280 is the control center of the electronic device, connecting various parts of the terminal through various interfaces and lines. It executes software programs stored in the memory 1220 and calls data stored in the memory 1220 to perform various functions and process data. In some embodiments, the processor 1280 may include one or more processing units; the processor 1280 may also integrate an application processor and a baseband processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the baseband processor mainly handles wireless communication. It is understood that the baseband processor may not be integrated into the processor 1280. In this application, the processor 1280 can run an operating system, applications, user interface display and touch response, and the methods of the embodiments of this application. Furthermore, the processor 1280 is coupled to the display unit 1230.

[0205] In this application embodiment, the aforementioned electronic device can be a smart mobile terminal, smart home device, tablet computer, laptop computer, smart handheld device, personal computer (PC), computer, smart screen, various wearable devices, personal digital assistant (PDA), etc. The electronic device can also be an independent physical server deploying user interaction functions, a server cluster or distributed system composed of multiple physical servers deploying user interaction functions, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms, and deploying fingerprint unlocking functions. This application does not limit the specific form of the aforementioned electronic device.

[0206] Based on the same inventive concept, embodiments of this application provide a computer-readable storage medium that, when the instructions in the storage medium are executed by a processor, enables the processor to perform any of the methods described in the above embodiments.

[0207] Based on the same inventive concept, this application also provides a computer program product, which includes computer program code that, when run on a computer, causes the computer to execute any of the methods described above. Since the methods for analyzing the subject evolution of the problems solved by the above computer program products are similar, the implementation of the above computer program products can be referred to the implementation of the methods, and repeated details will not be elaborated further.

[0208] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0209] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams.

[0210] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more processes in a flowchart and / or one or more blocks in a block diagram.

[0211] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more blocks in the block diagram.

[0212] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for thematic evolution analysis, characterized in that, include: For any historical time window among multiple consecutive historical time windows, a topic evolution analysis model is used to extract each topic in the corpus within the historical time window, and the first actual topic intensity of each topic in the historical time window is determined, wherein the size of the historical time window is obtained in advance by adjusting the window size using the topic evolution analysis model; For any given topic, based on the first actual topic intensity of the topic within each historical time window, the first predicted topic intensity of the topic within the current time window is determined. Based on each first predicted topic intensity and the second actual topic intensity of the topic within the current time window, the fusion value of the fusion time window corresponding to the topic is determined. Based on the fusion value, the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window are determined. Using the topic evolution analysis model after each hyperparameter is determined, topic extraction is performed on the corpus acquired within the current time window; The step of determining the fusion value of the fusion time window corresponding to the topic based on the intensity of each first predicted topic and the intensity of the topic in the second actual topic within the current time window includes: Based on the intensity of each first predicted topic and the intensity of the second actual topic within the current time window, the target deviation value corresponding to the topic is determined; If the target deviation value is not greater than the first threshold, then the sum of the fusion value corresponding to the previous historical time window of the current time window and the preset value is determined as the fusion value of the time window to be fused corresponding to the topic. If the target deviation value is not less than the second threshold, then the difference between the fusion value corresponding to the previous historical time window of the current time window and the preset value is determined as the fusion value of the time window to be fused corresponding to the topic. If the target deviation value is greater than the first threshold and less than the second threshold, then the fusion value corresponding to the previous historical time window of the current time window is determined as the fusion value of the time window to be fused corresponding to the topic.

2. The method as described in claim 1, characterized in that, The step of determining the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window based on the fusion value includes: Based on the fused values, determine the prior parameter weight matrix corresponding to the current time window; Based on the prior parameter weight matrix and the topic content evolution matrix corresponding to the current time window, determine the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window.

3. The method as described in claim 1, characterized in that, The step of determining the target deviation value corresponding to the topic based on the intensity of each first predicted topic and the intensity of the topic in the second actual topic within the current time window includes: Determine the absolute value of the difference between the intensity of each first predicted topic and the intensity of the second actual topic within the current time window; From all the absolute values, select the preset number of absolute values ​​that are ranked first in descending order; The target deviation value corresponding to the topic is determined based on the absolute value of the preset quantity.

4. The method as described in claim 1 or 2, characterized in that, The method further includes: The document-topic distribution for multiple consecutive target time windows is obtained using the aforementioned topic evolution analysis model, and a line graph of topic intensity corresponding to the multiple target time windows is plotted; and / or The topic evolution analysis model is used to obtain the topic-word distribution of multiple consecutive target time windows, and the topic content evolution diagrams corresponding to the multiple target time windows are drawn. Display the theme intensity line graph and / or the theme content evolution graph in the designated display interface.

5. The method as described in claim 1 or 2, characterized in that, The size of the historical time window is obtained as follows: For any initial time window among multiple consecutive initial time windows, the topic evolution analysis model is used to extract each topic in the corpus within the initial time window, and the third actual topic intensity of each topic in the initial time window is determined. The corpus within the initial time window is obtained by dividing the corpus within a preset time period based on window division rules. For any given topic, based on the third actual topic intensity within each initial time window, a second predicted topic intensity within the historical time window is determined, and based on the second predicted topic intensity and the fourth actual topic intensity within the historical time window, the deviation of the topic is determined; wherein, the historical time window is adjacent to and follows the plurality of initial time windows; The size of the historical time window is adjusted according to the deviation of each topic.

6. The method as described in claim 1 or 2, characterized in that, The topic evolution analysis model is an adaptive online topic evolution DAOLDA model, which is obtained by improving the online implicit Dirichlet distribution OLDA model.

7. A theme evolution analysis device, characterized in that, include: The first determining module is used to extract each topic in the corpus within any historical time window from multiple consecutive historical time windows using a topic evolution analysis model, and to determine the first actual topic intensity of each topic in the historical time window, wherein the size of the historical time window is obtained in advance by adjusting the window size using the topic evolution analysis model. The second determining module is used to determine, for any topic, the first predicted topic intensity of the topic in the current time window based on the first actual topic intensity of the topic in each historical time window, and to determine the fusion value of the time window to be fused corresponding to the topic based on each first predicted topic intensity and the second actual topic intensity of the topic in the current time window, and to determine the hyperparameters of the topic in the topic evolution analysis model corresponding to the current time window based on the fusion value. The topic extraction module is used to extract topics from the corpus acquired within the current time window using a topic evolution analysis model after each hyperparameter is determined. The step of determining the fusion value of the fusion time window corresponding to the topic based on the intensity of each first predicted topic and the intensity of the topic in the second actual topic within the current time window includes: Based on the intensity of each first predicted topic and the intensity of the second actual topic within the current time window, the target deviation value corresponding to the topic is determined; If the target deviation value is not greater than the first threshold, then the sum of the fusion value corresponding to the previous historical time window of the current time window and the preset value is determined as the fusion value of the time window to be fused corresponding to the topic. If the target deviation value is not less than the second threshold, then the difference between the fusion value corresponding to the previous historical time window of the current time window and the preset value is determined as the fusion value of the time window to be fused corresponding to the topic. If the target deviation value is greater than the first threshold and less than the second threshold, then the fusion value corresponding to the previous historical time window of the current time window is determined as the fusion value of the time window to be fused corresponding to the topic.

8. An electronic device, characterized in that, include: Memory is used to store computer programs that can be executed by the controller; The controller is connected to the memory and is configured to perform the method as described in any one of claims 1-6.

Images