Military report article generation method, device and equipment

Abstract

The invention discloses a military report article generation method, device and equipment, and the method comprises the steps: obtaining initial information in the military field, carrying out the preprocessing of the initial information including data cleaning, and obtaining a to-be-processed article; key information of the article to be processed is extracted through a pre-trained classification model and an extraction model, the classification model is obtained through training based on a TextCNN model, and the extraction model is obtained through training based on a TextRank algorithm; classifying the articles to be processed on the basis of the key information to obtain article sets corresponding to thematic types, calculating vector values of the articles in each article set, merging the articles with the vector values meeting a preset condition, and generating thematic report articles of the same semantic event; and integrating the thematic report articles according to a report template by utilizing a preset report large model to generate a military report article. The information processing efficiency and accuracy can be improved, and the generated military report is more professional, rapid and targeted.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method, device and equipment for generating militarized report articles. Background Art

[0002] In the field of military intelligence analysis, it is crucial to efficiently and accurately collect and process massive intelligence information from multiple different websites. However, the current research status faces a series of challenges. First, due to the extensiveness and diversity of intelligence sources, the collected data is diverse and complex, and inevitably contains a large amount of redundant information. These redundant data not only increase the storage burden, but also greatly increase the difficulty and cost of subsequent information extraction and collation. Although the traditional method of manually annotating training data can guarantee a certain degree of accuracy, it is inefficient, labor-intensive, and difficult to adapt to the strict requirements of timeliness and accuracy in military intelligence analysis.

[0003] On the other hand, with the development of artificial intelligence technology, zero-shot information extraction models have attracted much attention due to their characteristics of not requiring a large amount of labeled data. However, in the specific application scenario of military intelligence analysis, the performance of zero-shot models is often unsatisfactory. Due to the particularity of military intelligence data, such as high professionalism, sensitivity, and possible complex relationship networks, existing zero-shot models are difficult to accurately capture the core features and key information of the data in the absence of targeted training, resulting in low accuracy of the extraction results and failure to meet actual production needs. Summary of the invention

[0004] In view of this, the purpose of the present invention is to propose a method, device and equipment for generating militarized report articles, aiming to solve the problem of how to reduce the cost of information extraction and organization while improving the accuracy of information processing in military intelligence analysis to generate professional militarized reports.

[0005] To achieve the above object, the present invention provides a method for generating a militarized report article, the method comprising:

[0006] Acquire initial information in the military field, and perform preprocessing including data cleaning on the initial information to obtain articles to be processed;

[0007] Extracting key information of the article to be processed using a pre-trained classification model and an extraction model, wherein the classification model is trained based on a TextCNN model, and the extraction model is trained based on a TextRank algorithm;

[0008] Classifying the articles to be processed based on the key information to obtain a set of articles of corresponding topic types, calculating the vector value of each article in each set of articles, merging the articles whose vector values ​​meet the preset conditions, and generating topic report articles of the same semantic event;

[0009] The preset report model is used to integrate the special report articles according to the report template to generate a militarized report article.

[0010] Preferably, the method further comprises:

[0011] User information including user browsing habits is obtained to determine the user portrait, matching topic types are screened out according to the user portrait, and the topic report articles corresponding to the topic types are input into the report model to generate a customized topic report for push.

[0012] Preferably, the key information includes article categories, military keywords and military entities; the method of classifying the articles to be processed based on the key information to obtain a set of articles of corresponding topic types includes:

[0013] Calculating the weight values ​​of the article category, the military keyword, and the military entity respectively to obtain the classification relevance weight, the keyword relevance weight, and the entity relevance weight;

[0014] Calculate the corresponding weighted averages according to the classification relevance weight, keyword relevance weight and entity relevance weight to obtain the classification weighted average, entity weighted average and keyword weighted average;

[0015] The average of the classification weighted average, the entity weighted average and the keyword weighted average is calculated to obtain a combined weighted average, and the articles to be processed whose combined weighted average is greater than a combined threshold are regarded as the article set of the same topic type.

[0016] Preferably, respectively calculating the weight values ​​of the article category, the military keyword, and the military entity to obtain the classification association weight, the keyword association weight, and the entity association weight includes:

[0017] The classification association weight WTag(i) is calculated using WTag(i)=PTag(i)*a, where PTag(i) represents the accuracy of the classification model when extracting the article category, and a represents a fixed coefficient;

[0018] The entity association weight WEntity(i) is calculated using WEntity(i)=W(Vi)*PEntity(i), wherein W(Vi) represents the weight of the military entity, and PEntity(i) represents the frequency of occurrence of the military entity;

[0019] The keyword association weight WKey(i) is calculated using WKey(i)=W'(Vi)*PKey(i), wherein W'(Vi) represents the weight of the military keyword, and PKey(i) represents the frequency of occurrence of the military keyword.

[0020] Preferably, the corresponding weighted average values ​​are calculated according to the classification relevance weights, keyword relevance weights and entity relevance weights to obtain the classification weighted average value, the entity weighted average value and the keyword weighted average value, including:

[0021] The weighted average value of the classification is obtained by calculating using the formula ∑(Ma*h(WTag(i)>ma) / ∑Ma;

[0022] The weighted average value of the entity is obtained by calculating using the formula ∑(Mo*h(WEntity(i)>mo)>M) / ∑Mo;

[0023] The weighted average value of the keywords is calculated using the formula ∑(Md*h(WKey(i)>md) / ∑Md; wherein Ma represents the preset classification weighted value, Mo represents the preset entity similarity weighted value, Md represents the preset keyword weighted value, ma represents the classification filtering threshold, mo represents the entity filtering threshold, md represents the keyword filtering threshold, M represents the merging threshold, h function represents the filtering condition that needs to be met in the weighting process, WTag(i) represents the classification association weight, WEntity(i) represents the entity association weight, and WKey(i) represents the keyword association weight.

[0024] Preferably, after merging the articles whose vector values ​​meet the preset conditions to generate special report articles of the same semantic event, the method further includes:

[0025] The classification association weight, the keyword association weight and the entity association weight are integrated by using a preset f function to obtain a total association, and the similarity of the articles in the special report article is judged according to the total association, wherein the f function is defined as: f = (WTag(i-1)*Ma+WEntity(i-1)*Mo+WKey(i-1)*Md) / N, and N represents the total number of the classification association weight WTag(i) of the article category, the entity association weight WEntity(i) of the military entity, and the keyword association weight WKey(i) of the military keyword in the current article node.

[0026] Preferably, the use of a preset report macromodel to integrate the special report articles according to a report template to generate a militarized report article includes:

[0027] Determine the target article in the special report article, and form the event context of the target article according to the event content and time relationship;

[0028] Based on the Promote constructed by the user portrait, the content corresponding to the event context is filled into the report template through the report macro model to obtain the militarized report article.

[0029] To achieve the above-mentioned purpose, the present invention also provides a device for generating a militarized report article, the device comprising:

[0030] A preprocessing unit, used to obtain initial information in the military field, and perform preprocessing including data cleaning on the initial information to obtain articles to be processed;

[0031] An extraction unit, used to extract key information of the article to be processed by using a pre-trained classification model and an extraction model, wherein the classification model is trained based on a TextCNN model, and the extraction model is trained based on a TextRank algorithm;

[0032] A classification unit, used for classifying the articles to be processed based on the key information to obtain a set of articles of corresponding topic types, and merging articles whose vector values ​​meet preset conditions by calculating the vector values ​​of each article in each set of articles to generate topic report articles of the same semantic event;

[0033] The generating unit is used to integrate the special report articles according to the report template by using the preset report model to generate the militarized report articles.

[0034] In order to achieve the above objectives, the present invention also proposes a militarized report article generation device, including a processor, a memory, and a computer program stored in the memory, wherein the computer program is executed by the processor to implement the steps of a militarized report article generation method as described in the above embodiment.

[0035] In order to achieve the above objectives, the present invention also proposes a computer-readable storage medium, on which a computer program is stored. The computer program is executed by a processor to implement the steps of a method for generating a militarized report article as described in the above embodiment.

[0036] In order to achieve the above objectives, the present invention also proposes a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of a method for generating a militarized report article as described in the above embodiment.

[0037] Beneficial effects:

[0038] The above scheme extracts key information from the initial information in the military field, and classifies the articles based on this key information to generate special reports, and finally uses the report model to integrate and generate militarized report articles. This process significantly reduces the time and cost of manual processing, improves the efficiency and accuracy of information processing, and makes the generated militarized reports more professional, fast and targeted.

[0039] The above solution, by introducing user portraits and customized push mechanisms, can filter and push special reports that match users based on their browsing habits and interest preferences; this personalized service not only improves the user experience, but also enhances the pertinence and effectiveness of information, allowing users to obtain the military information they care about more quickly.

[0040] The above scheme calculates the weights of article categories, military keywords and military entities, and uses these weights for classification and generation of special reports. By comprehensively considering the correlation weights of multiple dimensions, it can more accurately judge the semantic similarity and special topic affiliation of articles, thereby generating more accurate and valuable special reports.

[0041] The above scheme, through the process of calculating classification relevance weight, entity relevance weight and keyword relevance weight, can effectively quantify the importance of different factors in the classification and report generation process, and provide a reliable basis for subsequent weighted average calculation and classification decision.

[0042] The above scheme, by calculating the weighted average of classification, the weighted average of entities and the weighted average of keywords and setting corresponding filtering thresholds, can more scientifically evaluate the similarity between the articles to be processed and the specific topics, help reduce the errors caused by subjective judgment, and improve the accuracy of classification and integration.

[0043] The above scheme, through the process of integrating the event context into a militarized report article by using the report macro model and report template, can organize scattered special articles into an orderly and logical report, thereby improving the readability and practicality of the report; at the same time, using the report macro model for content filling can also ensure the accuracy and professionalism of the report content. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1A flowchart of a method for generating a militarized report article provided in accordance with an embodiment of the present invention.

[0046] Figure 2 A schematic diagram of the structure of a militarized report article generating device provided in one embodiment of the present invention.

[0047] The realization of the purpose, functional features and advantages of the invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0049] The content of the present invention is described in detail below with reference to embodiments.

[0050] Reference Figure 1 The figure is a flow chart of a method for generating a militarized report article provided in one embodiment of the present invention.

[0051] In this embodiment, the method includes:

[0052] S11, obtaining initial information in the military field, and performing preprocessing including data cleaning on the initial information to obtain articles to be processed.

[0053] In this embodiment, crawler frameworks such as crawllab are used to collect initial military information from different target websites in a targeted manner, and the initial information is preprocessed (including removing messy labels, abnormal characters, meaningless data, etc.) to ensure the accuracy and normalization of the data.

[0054] S12, using a pre-trained classification model and an extraction model to extract key information of the article to be processed, wherein the classification model is trained based on a TextCNN model, and the extraction model is trained based on a TextRank algorithm.

[0055] In this embodiment, after fine-tuning training by selecting a suitable model and training data, the model obtained by the training is used to extract key information including military entities, military keywords, article categories, etc. from the data obtained by the crawler. Specifically, the accumulated military data (equipment data, location data, organization data, military data, country and leader data, and project data stored in the database, etc.) are used to train the TextCNN model (whose network structure includes a layer of convolution, a layer of max-pooling, and finally the output is externally connected to softmax for n classification) to obtain a classification model, and the classification model is used to extract the article category of the article to be processed (by filtering the important features in the article to be processed, and using the feature map extracted by the convolution layer to output to the article category). The military terms accumulated in the vocabulary (such as the US Air Force, aircraft, warships, and the Joint Anti-Small Unmanned Aerial Systems Office, etc.) are used to extract military keywords in the article to be processed through the extraction model trained by the TextRank algorithm. The military terms accumulated in the vocabulary (such as military equipment names, place names, military figures, etc.) are used to extract military entities through the extraction model trained by the TextRank algorithm.

[0056] S13, classifying the articles to be processed based on the key information to obtain a set of articles of corresponding topic types, calculating the vector value of each article in each article set, merging the articles whose vector values ​​meet the preset conditions, and generating topic report articles of the same semantic event.

[0057] Furthermore, the key information includes article categories, military keywords and military entities; in step S13, the articles to be processed are classified based on the key information to obtain a set of articles of corresponding topic types, including:

[0058] S13-1, respectively calculating the weight values ​​of the article category, the military keyword and the military entity to obtain the classification association weight, the keyword association weight and the entity association weight.

[0059] Furthermore, in step S13-1, the weight values ​​of the article category, the military keyword and the military entity are calculated respectively to obtain the classification association weight, the keyword association weight and the entity association weight, including:

[0060] The classification association weight WTag(i) is calculated using WTag(i)=PTag(i)*a, where PTag(i) represents the accuracy of the classification model when extracting the article category, and a represents a fixed coefficient;

[0061] The entity association weight WEntity(i) is calculated using WEntity(i)=W(Vi)*PEntity(i), wherein W(Vi) represents the weight of the military entity, and PEntity(i) represents the frequency of occurrence of the military entity;

[0062] The keyword association weight WKey(i) is calculated using WKey(i)=W'(Vi)*PKey(i), wherein W'(Vi) represents the weight of the military keyword, and PKey(i) represents the frequency of occurrence of the military keyword.

[0063] S13-2, calculating the corresponding weighted averages according to the classification relevance weights, keyword relevance weights and entity relevance weights, and obtaining the classification weighted average, entity weighted average and keyword weighted average.

[0064] Furthermore, in step S13-2, the corresponding weighted averages are calculated according to the classification relevance weights, keyword relevance weights and entity relevance weights to obtain the classification weighted averages, entity weighted averages and keyword weighted averages, including:

[0065] The weighted average value of the classification is obtained by calculating using the formula ∑(Ma*h(WTag(i)>ma) / ∑Ma;

[0066] The weighted average value of the entity is obtained by calculating using the formula ∑(Mo*h(WEntity(i)>mo)>M) / ∑Mo;

[0067] The weighted average value of the keywords is calculated using the formula ∑(Md*h(WKey(i)>md) / ∑Md; wherein Ma represents the preset classification weighted value, Mo represents the preset entity similarity weighted value, Md represents the preset keyword weighted value, ma represents the classification filtering threshold, mo represents the entity filtering threshold, md represents the keyword filtering threshold, M represents the merging threshold, h function represents the filtering condition that needs to be met in the weighting process, WTag(i) represents the classification association weight, WEntity(i) represents the entity association weight, and WKey(i) represents the keyword association weight.

[0068] S13-3, calculating the average of the classification weighted average, the entity weighted average and the keyword weighted average to obtain a combined weighted average, and treating the articles to be processed whose combined weighted average is greater than the combined threshold as the article set of the same topic type.

[0069] Furthermore, after merging the articles whose vector values ​​meet the preset conditions to generate special report articles of the same semantic event, the method further includes:

[0070] The classification association weight, the keyword association weight and the entity association weight are integrated by using a preset f function to obtain a total association, and the similarity of the articles in the special report article is judged according to the total association, wherein the f function is defined as: f = (WTag(i-1)*Ma+WEntity(i-1)*Mo+WKey(i-1)*Md) / N, and N represents the total number of the classification association weight WTag(i) of the article category, the entity association weight WEntity(i) of the military entity, and the keyword association weight WKey(i) of the military keyword in the current article node.

[0071] In this embodiment, based on military data, the classification label Tag(i) (wherein the classification label is the article category) and its accuracy PTag(i) extracted by the classification model obtained by training the TextCNN model in a small batch size are used to calculate its classification association weight WTag(i) (the classification association weight is defined as the accuracy PTag(i) of the article category based on its training data as the association weight of the document, multiplied by a fixed coefficient and taken as an integer to increase the weight of the label in the subsequent entire calculation process, such as setting a fixed coefficient a=10), that is, WTag(i)=PTag(i)*a.

[0072] Based on military data, the weight W(Vi) of military entity i and the frequency of occurrence of military entity PEntity(i) extracted by the small-scale training model TextRank are multiplied to determine the entity relevance weight WEntity(i), that is, WEntity(i) = W(Vi)*PEntity(i); where military entity i represents the i-th military entity in the article, and WEntity(i) represents the weight of the i-th military entity in the entire article. Where W(Vi) represents the weight of node Vi, d represents the damping factor (usually set to 0.85), which is used to simulate the probability of random jumps to other nodes, In(Vi) represents the set of all nodes pointing to node Vi, Out(Vj) represents the number of all nodes pointed to by node Vj; where node refers to military entities, and the probability of jumps between military entities is calculated, and Vi and Vj represent the probability of jumps between any two different nodes.

[0073] Based on military data, the weight W'(Vi) of the military keyword Key(i) extracted by the small-scale training model TextRank (W'(Vi) calculation reference ) and the frequency of occurrence of military keywords, and the keyword relevance weight WKey(i) is determined by multiplying them, that is, WKey(i)=W'(Vi)*PKey(i);

[0074] By calculating the weights extracted from each dimension (dimensions include article categories, military keywords, military entities, etc.), the information is aggregated and processed, and the features of the same article are fused. Specifically, through multiple fusion processes of various features, the data features can capture higher-level semantics, and select features with relatively high weights as key features (for example, an article can extract relevant weights from different angles (such as article categories, military keywords, etc.), but each weight is only valid for the current dimension. When there are two articles A and B for comparison, the weight of the article category dimension of A is higher than that of B, and the weight of the military keyword dimension of B is higher than that of A. Finally, it is difficult to compare which one has a higher weight between articles A and B, so it is necessary to abstract to a higher dimension to calculate the correlation of the total dimensions when there are different dimensions in the article), as follows;

[0075] Its feature update function can be expressed as Where T i+1 represents the feature representation in the i+1 layer, N represents the total number of the classification association weight WTag(i) of the article category, the entity association weight WEntity(i) of the military entity, and the keyword association weight WKey(i) of the military keyword in the current document node (where i represents the i-th classification label or military entity or military keyword extracted and processed in the article), T i u represents the weighted average of the i-th (military keyword, article category, military entity set) in the current document, and f is an aggregation function. The f function is defined as: f = (WTag (i-1) * Ma + WEntity (i-1) * Mo + WKey (i-1) * Md) / N (N is the total number).

[0076] All articles in a certain period of time are used to calculate the total relevance value of each article. According to the total relevance value, it is convenient to judge the similarity of articles. When the user portrait (including his topic of interest) is obtained, the articles in the topic report may contain articles about the same event. When the total relevance between articles is very close, it means that the articles are very similar. At this time, the articles need to be filtered. The filtering method is: by comparing the relevance between articles (for example, the difference is within 0.1), it is considered to be the same article, and the duplicates should be removed (any one can be selected) and then integrated and output to the large model.

[0077] The topic content is further determined by calculating the weighted average of each article in the topic:

[0078] Classification weighted average = ∑(Ma*h(WTag(i)>ma) / ∑Ma,

[0079] Entity weighted average = ∑(Mo*h(WEntity(i)>mo)>M) / ∑Mo,

[0080] Weighted average value of keywords = ∑(Md*h(WKey(i)>md) / ∑Md,

[0081] Among them, the filtering thresholds are set by setting the classification weight value Ma, entity similarity weight value Mo, and keyword weight value Md: classification filtering threshold ma, entity filtering threshold mo, entity filtering threshold md, and merge threshold M; the h function represents the conditional rules for filtering in the weighted process, such as the condition WTag(i)-ma>0 must be met (that is, the corresponding classification association weight in the extracted classification labels, select the classification labels greater than ma).

[0082] Finally, the combined weighted average value Tu = (classification weighted average value + entity weighted average value + keyword weighted average value) / 3 is calculated, and all the articles to be processed whose combined weighted average value Tu is greater than the combined threshold value M are regarded as a set of articles of the same topic type.

[0083] Document content vectorization: First, use the TF-IDF algorithm to construct a vector matrix of the article collection content (that is, the vector matrix obtained after vectorizing each article), where each row represents a document (article), and each column represents the TF-IDF vector value of a word. Use the cosine similarity method to compare similar vector values. By comparing the TF-IDF value difference of the above content with the feature value of the above article, according to the threshold you set, when both conditions are met at the same time, you can determine whether the two articles can be merged into event content articles with the same semantics.

[0084] S14, using a preset report model to integrate the special report articles according to a report template to generate a militarized report article.

[0085] Furthermore, in step S14, the preset report model is used to integrate the special report articles according to the report template to generate a militarized report article, including:

[0086] S14-1, determining target articles in the special report articles, and forming event contexts for the target articles according to event content and time relationships;

[0087] S14-2, based on the Promote constructed by the user portrait, the content corresponding to the event context is filled into the report template through the report macro model to obtain the militarization report article.

[0088] In this embodiment, the same and similar articles are classified as special report articles of the same semantic event through the above classification, and then the military entities, event time, etc. of each article are extracted. The similarity between each article is judged according to the above process, and the weighted average of the contents that are too similar is taken as the first content. According to the event content, time relationship, etc., they are merged and organized into an event context.

[0089] Based on the existing annotated professional data in the military professional field, the model is trained and fine-tuned to generate a large report model. Based on the user portrait, promote is built and integrated, and the relevant event article content and event context data are filled into the constructed report template to form a complete militarized report article.

[0090] Furthermore, the method further comprises:

[0091] S15, obtaining user information including user browsing habits to determine the user portrait, screening out matching topic types according to the user portrait, inputting the topic report articles corresponding to the topic types into the report macro model to generate a customized topic report for push.

[0092] In this embodiment, by collecting label data of user portraits including user interest areas, behavioral preferences, historical browsing records, etc., the user's classification label is defined according to the user's past access record habits, and the article classification labels corresponding to the browsed articles based on the classification labels of other users are used as the data for original corpus training to train the large model. The large model can recommend and locate the relevant type of special report content that the corresponding user is interested in in the huge data through user portrait labels to generate customized special reports.

[0093] Reference Figure 2 The figure is a schematic diagram of the structure of a militarized report article generating device provided by one embodiment of the present invention.

[0094] In this embodiment, the device 20 includes:

[0095] A preprocessing unit 21 is used to obtain initial information in the military field and perform preprocessing including data cleaning on the initial information to obtain articles to be processed;

[0096] An extraction unit 22 is used to extract key information of the article to be processed by using a pre-trained classification model and an extraction model, wherein the classification model is trained based on a TextCNN model, and the extraction model is trained based on a TextRank algorithm;

[0097] The classification unit 23 is used to classify the articles to be processed based on the key information to obtain a set of articles of corresponding topic types, calculate the vector value of each article in each set of articles, merge the articles whose vector values ​​meet the preset conditions, and generate topic report articles of the same semantic event;

[0098] The generating unit 24 is used to integrate the special report articles according to the report template by using the preset report macro model to generate a militarized report article.

[0099] In another embodiment, the device 20 further comprises:

[0100] The customization unit 25 is used to obtain user information including user browsing habits to determine the user portrait, filter out matching topic types according to the user portrait, input the topic report articles corresponding to the topic types into the report macro model to generate a customized topic report for push.

[0101] Each unit module of the device 20 can respectively execute the corresponding steps in the above method embodiment, so each unit module will not be described in detail here. Please refer to the description of the above corresponding steps for details.

[0102] The embodiment of the present invention also provides a militarization report article generation device, the device includes the militarization report article generation device as described above, wherein the militarization report article generation device can adopt Figure 2 The structure of the embodiment can be executed accordingly. Figure 1 The technical solution of the method embodiment shown has similar implementation principles and technical effects. For details, please refer to the relevant records in the above embodiments, which will not be repeated here.

[0103] The device includes: a mobile phone, a digital camera, a tablet computer, or other devices with a camera function, or a device with an image processing function, or a device with an image display function. The device may include components such as a memory, a processor, an input unit, a display unit, and a power supply.

[0104] Among them, the memory can be used to store software programs and modules, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as an image playback function, etc.), etc.; the data storage area may store data created according to the use of the device, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. Accordingly, the memory may also include a memory controller to provide the processor and the input unit with access to the memory.

[0105] The input unit can be used to receive input digital or character or image information, and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control. Specifically, the input unit of this embodiment includes not only a camera, but also a touch-sensitive surface (such as a touch display) and other input devices.

[0106] The display unit can be used to display information input by the user or information provided to the user and various graphical user interfaces of the device, which can be composed of graphics, text, icons, videos and any combination thereof. The display unit may include a display panel, and optionally, the display panel may be configured in the form of LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), etc. Further, the touch-sensitive surface may cover the display panel, and when the touch-sensitive surface detects a touch operation on or near it, it is transmitted to the processor to determine the type of touch event, and then the processor provides a corresponding visual output on the display panel according to the type of touch event.

[0107] The embodiment of the present invention further provides a computer-readable storage medium, which may be a computer-readable storage medium included in the memory in the above embodiment; or a computer-readable storage medium that exists independently and is not installed in a device. The computer-readable storage medium stores at least one instruction, which is loaded and executed by a processor to implement Figure 1 The method for generating militarized report articles is shown. The computer-readable storage medium may be a read-only memory, a disk or an optical disk, etc.

[0108] The embodiment of the present invention also provides a computer program product, including a computer program / instruction, which is loaded and executed by a processor to implement Figure 1 A method for generating a militarized report article is shown.

[0109] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, equipment embodiment and storage medium embodiment, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0110] Furthermore, in this document, the terms "comprises," "comprising," or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device that includes the element.

[0111] The above description shows and describes the preferred embodiments of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the invention, through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art shall not depart from the spirit and scope of the present invention, and shall be within the scope of protection of the claims attached to the present invention.

Claims

1. A method for generating a militarized report article, characterized in that: The method comprises: Acquire initial information in the military field, and perform preprocessing including data cleaning on the initial information to obtain articles to be processed; Extracting key information of the article to be processed using a pre-trained classification model and an extraction model, wherein the classification model is trained based on a TextCNN model, and the extraction model is trained based on a TextRank algorithm; Classifying the articles to be processed based on the key information to obtain a set of articles of corresponding topic types, calculating the vector value of each article in each set of articles, merging the articles whose vector values ​​meet the preset conditions, and generating topic report articles of the same semantic event; The preset report model is used to integrate the special report articles according to the report template to generate a militarized report article.

2. A method for generating militarized report articles according to claim 1, characterized in that: The method further comprises: User information including user browsing habits is obtained to determine the user portrait, matching topic types are screened out according to the user portrait, and the topic report articles corresponding to the topic types are input into the report model to generate a customized topic report for push.

3. A method for generating militarized report articles according to claim 1, characterized in that: The key information includes article categories, military keywords and military entities; the articles to be processed are classified based on the key information to obtain a set of articles of corresponding topic types, including: Calculating the weight values ​​of the article category, the military keyword, and the military entity respectively to obtain the classification relevance weight, the keyword relevance weight, and the entity relevance weight; Calculate the corresponding weighted averages according to the classification relevance weight, keyword relevance weight and entity relevance weight to obtain the classification weighted average, entity weighted average and keyword weighted average; The average of the classification weighted average, the entity weighted average and the keyword weighted average is calculated to obtain a combined weighted average, and the articles to be processed whose combined weighted average is greater than a combined threshold are regarded as the article set of the same topic type.

4. A method for generating militarized report articles according to claim 3, characterized in that: The step of respectively calculating the weight values ​​of the article category, the military keyword, and the military entity to obtain the classification relevance weight, the keyword relevance weight, and the entity relevance weight includes: The classification association weight WTag(i) is calculated using WTag(i)=PTag(i)*a, where PTag(i) represents the accuracy of the classification model when extracting the article category, and a represents a fixed coefficient; The entity association weight WEntity(i) is calculated using WEntity(i)=W(Vi)*PEntity(i), wherein W(Vi) represents the weight of the military entity, and PEntity(i) represents the frequency of occurrence of the military entity; The keyword association weight WKey(i) is calculated using WKey(i)=W'(Vi)*PKey(i), wherein W'(Vi) represents the weight of the military keyword, and PKey(i) represents the frequency of occurrence of the military keyword.

5. A method for generating militarized report articles according to claim 3, characterized in that: The corresponding weighted average values ​​are calculated according to the classification relevance weights, keyword relevance weights and entity relevance weights to obtain the classification weighted average value, entity weighted average value and keyword weighted average value, including: The weighted average value of the classification is obtained by calculating using the formula ∑(Ma*h(WTag(i)>ma) / ∑Ma; The weighted average value of the entity is obtained by calculating using the formula ∑(Mo*h(WEntity(i)>mo)>M) / ∑Mo; The weighted average value of the keywords is calculated using the formula ∑(Md*h(WKey(i)>md) / ∑Md; wherein Ma represents the preset classification weighted value, Mo represents the preset entity similarity weighted value, Md represents the preset keyword weighted value, ma represents the classification filtering threshold, mo represents the entity filtering threshold, md represents the keyword filtering threshold, M represents the merging threshold, h function represents the filtering condition that needs to be met in the weighting process, WTag(i) represents the classification association weight, WEntity(i) represents the entity association weight, and WKey(i) represents the keyword association weight.

6. A method for generating militarized report articles according to claim 5, characterized in that: After merging the articles whose vector values ​​meet the preset conditions to generate special report articles of the same semantic event, the method further includes: The classification association weight, the keyword association weight and the entity association weight are integrated by using a preset f function to obtain a total association, and the similarity of the articles in the special report article is judged according to the total association, wherein the f function is defined as: f = (WTag(i-1)*Ma+WEntity(i-1)*Mo+WKey(i-1)*Md) / N, and N represents the total number of the classification association weight WTag(i) of the article category, the entity association weight WEntity(i) of the military entity, and the keyword association weight WKey(i) of the military keyword in the current article node.

7. A method for generating militarized report articles according to claim 1, characterized in that: The preset report model is used to integrate the special report articles according to the report template to generate a militarized report article, including: Determine the target article in the special report article, and form the event context of the target article according to the event content and time relationship; Based on the Promote constructed by the user portrait, the content corresponding to the event context is filled into the report template through the report macro model to obtain the militarized report article.

8. A device for generating militarized report articles, characterized in that: The device comprises: A preprocessing unit, used to obtain initial information in the military field, and perform preprocessing including data cleaning on the initial information to obtain articles to be processed; An extraction unit, used to extract key information of the article to be processed by using a pre-trained classification model and an extraction model, wherein the classification model is trained based on a TextCNN model, and the extraction model is trained based on a TextRank algorithm; A classification unit, used for classifying the articles to be processed based on the key information to obtain a set of articles of corresponding topic types, and merging articles whose vector values ​​meet preset conditions by calculating the vector values ​​of each article in each set of articles to generate topic report articles of the same semantic event; The generating unit is used to integrate the special report articles according to the report template by using the preset report model to generate the militarized report articles.

9. A military report article generating device, characterized in that: The invention comprises a processor, a memory and a computer program stored in the memory, wherein the computer program is executed by the processor to implement the steps of a method for generating a militarized report article as claimed in any one of claims 1 to 7.

10. A computer program product, comprising a computer program / instruction, wherein the computer program / instruction is executed by a processor to implement the steps of a method for generating a militarized report article as claimed in any one of claims 1 to 7.

Images