A method and device for generating article outline and a device for generating article outline

Abstract

The embodiments of the present invention provide a method and device for generating an article outline, and a device for generating an article outline. The method comprises: obtaining at least two reference articles that meet a preset theme and determining the paragraphs of each reference article; determining the candidate titles corresponding to each paragraph of each reference article; clustering the candidate titles corresponding to each paragraph of the at least two reference articles to obtain a global title connection graph corresponding to the preset theme; and generating an article outline corresponding to the preset theme according to the global title connection graph. The article outline generated by the present invention can fully cover the information related to the preset theme in multiple articles that meet the preset theme, thereby improving the relevance of the generated article outline to the preset theme; and, by generating an article outline based on the paragraph titles of multiple reference articles, the present invention enhances the credibility of the generated article outline, which is conducive to improving the writing assistance effect.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and device for generating an article outline, and a device for generating an article outline. Background Art

[0002] With the vigorous development of mobile Internet technology, massive amounts of information are gathered in various information flow platforms. The huge number of articles provides readers with almost endless choices. However, for article producers, the threshold for writing and publishing has been lowered, but it is becoming increasingly difficult to stand out from the massive amount of articles.

[0003] Currently, segmentation models are usually trained based on the outline text of Wikipedia and papers. The first-level segments and paragraph titles of the article are determined based on the trained segmentation model, and an article outline is generated based on the paragraph titles and recommended to users to assist them in writing. Summary of the invention

[0004] The embodiments of the present invention provide an article outline generation method, a device and a device for generating an article outline, which can improve the relevance of the generated article outline with a preset theme and enhance the credibility of the article outline.

[0005] In order to solve the above problem, an embodiment of the present invention discloses a method for generating an article outline, the method comprising:

[0006] Obtain at least two reference articles that match the pre-set topic and identify the paragraphs of each reference article;

[0007] Determine candidate titles for each paragraph of each reference article;

[0008] Performing clustering processing on the candidate titles corresponding to each paragraph of the at least two reference articles to obtain a global title connection graph corresponding to the preset topic;

[0009] An article outline corresponding to the preset topic is generated according to the global title connection diagram.

[0010] Optionally, clustering the candidate titles corresponding to each paragraph of the at least two reference articles to obtain a global title connection graph corresponding to the preset topic includes:

[0011] Merge the candidate titles corresponding to each paragraph to obtain the target title corresponding to each paragraph;

[0012] Connecting the target titles corresponding to the paragraphs of the same reference article in sequence according to the order of the paragraphs to obtain at least two internal title connection graphs, each of which corresponds to a reference article;

[0013] Clustering is performed on each target title in the at least two internal title connection graphs according to the semantics of the target title to obtain a global title connection graph corresponding to the preset theme.

[0014] Optionally, clustering the target titles in the at least two internal title connection graphs according to the semantics of the target titles to obtain a global title connection graph corresponding to the preset topic includes:

[0015] Clustering the target titles belonging to different reference articles according to the semantics of the target title to obtain the cluster to which each target title belongs;

[0016] The target titles belonging to the same cluster in the at least two internal title connection graphs are connected to obtain a global title connection graph corresponding to the preset theme.

[0017] Optionally, the step of connecting the target titles belonging to the same cluster in the at least two internal title connection graphs to obtain a global title connection graph corresponding to the preset theme includes:

[0018] Determining whether there are at least two identical target titles in the same cluster of the at least two internal title connection graphs;

[0019] If there are at least two identical target titles in the same cluster, the at least two identical target titles are merged to obtain a merged target title, and the target title in the cluster to which it belongs is updated according to the merged target title;

[0020] The updated target titles in the same cluster of the at least two internal title connection graphs are connected to obtain a global title connection graph corresponding to the preset theme.

[0021] Optionally, generating an article outline corresponding to the preset topic according to the global title connection diagram includes:

[0022] Extracting a target subgraph in the global title connection graph based on a community discovery algorithm;

[0023] Generate an article outline corresponding to the preset topic according to the target subgraph.

[0024] Optionally, the candidate titles include a first candidate title and a second candidate title, and determining the candidate title corresponding to each paragraph of each reference article includes:

[0025] For each paragraph of each reference article, generate at least two paragraph headings;

[0026] Determine a first candidate title from each paragraph title of each paragraph, wherein the similarity between the first candidate title and other paragraph titles of the paragraph to which it belongs satisfies a first preset condition;

[0027] A second candidate title is determined, the similarity of which with the first candidate title satisfies a second preset condition.

[0028] Optionally, determining the first candidate title from each paragraph title of each paragraph includes:

[0029] Calculating the sum of similarities between the first paragraph title and other paragraph titles of the same paragraph to obtain a correlation index of the first paragraph title, where the first paragraph title is any one of at least two paragraph titles of the paragraph to which it belongs;

[0030] If the relevance index of the first paragraph title is greater than the relevance indexes of other paragraph titles of the paragraph to which it belongs, the first paragraph title is determined to be the first candidate title.

[0031] Optionally, the second preset condition includes: the similarity between the second candidate title and the first candidate title is less than the similarity between the second paragraph title and the first candidate title, and the second paragraph title is any paragraph title that is not a candidate title among at least two paragraph titles of the paragraph to which the first candidate title belongs.

[0032] Optionally, obtaining at least two reference articles that match a preset topic and determining a paragraph of each reference article includes:

[0033] Receive a preset topic and screening conditions input by a user, wherein the screening conditions include at least one of article type, number of citations, publication time, article reading volume, and article like volume;

[0034] Searching for at least two reference articles that meet the preset topic according to the screening condition;

[0035] The paragraphs of each reference article are determined based on a pre-trained segmentation model.

[0036] Optionally, before determining the paragraphs of each reference article based on the pre-trained segmentation model, the method further includes:

[0037] Construct a segmentation model based on a bidirectional autoencoder model and conditional random fields;

[0038] Acquire segmented training corpus, wherein the segmented training corpus includes sample articles and segmentation points corresponding to the sample articles;

[0039] The segmentation model is trained according to the segmentation training corpus until the segmentation model meets a convergence condition to obtain a trained segmentation model.

[0040] On the other hand, an embodiment of the present invention discloses a device for generating an article outline, the device comprising:

[0041] A reference article acquisition module, used to acquire at least two reference articles that meet the preset theme and determine the paragraphs of each reference article;

[0042] A candidate title determination module is used to determine a candidate title corresponding to each paragraph of each reference article;

[0043] A title connection graph generating module, used for clustering the candidate titles corresponding to each paragraph of the at least two reference articles to obtain a global title connection graph corresponding to the preset topic;

[0044] An article outline generation module is used to generate an article outline corresponding to the preset theme according to the global title connection diagram.

[0045] Optionally, the title connection diagram generating module includes:

[0046] The title merging submodule is used to merge the candidate titles corresponding to each paragraph to obtain the target title corresponding to each paragraph;

[0047] An internal title connection graph generation submodule is used to sequentially connect the target titles corresponding to the paragraphs of the same reference article according to the order of the paragraphs to obtain at least two internal title connection graphs, each of which corresponds to a reference article;

[0048] The global title connection graph generating submodule is used to perform clustering processing on each target title in the at least two internal title connection graphs according to the semantics of the target title to obtain a global title connection graph corresponding to the preset theme.

[0049] Optionally, the global title connection graph generation submodule includes:

[0050] A clustering processing unit, used for clustering target titles belonging to different reference articles according to the semantics of the target titles to obtain the cluster to which each target title belongs;

[0051] The global title connection graph generating unit is used to connect the target titles belonging to the same cluster in the at least two internal title connection graphs to obtain the global title connection graph corresponding to the preset theme.

[0052] Optionally, the global title connection graph generating unit includes:

[0053] a target title determination subunit, used to determine whether there are at least two identical target titles in the same cluster of the at least two internal title connection graphs;

[0054] A target title merging subunit is used for merging at least two identical target titles to obtain a merged target title if there are at least two identical target titles in the same cluster, and updating the target title in the cluster to which it belongs according to the merged target title;

[0055] The target title connection subunit is used to connect the updated target titles in the same cluster of the at least two internal title connection graphs to obtain a global title connection graph corresponding to the preset theme.

[0056] Optionally, the article outline generating module includes:

[0057] A target subgraph extraction submodule, used for extracting a target subgraph from the global title connection graph based on a community discovery algorithm;

[0058] The article outline generation submodule is used to generate an article outline corresponding to the preset theme according to the target subgraph.

[0059] Optionally, the candidate title includes a first candidate title and a second candidate title, and the candidate title determination module includes:

[0060] A paragraph title generation submodule, used for generating at least two paragraph titles for each paragraph of each reference article;

[0061] A first candidate title determination submodule, used to determine a first candidate title from each paragraph title of each paragraph, wherein the similarity between the first candidate title and other paragraph titles of the paragraph to which it belongs meets a first preset condition;

[0062] The second candidate title determination submodule is used to determine a second candidate title whose similarity with the first candidate title meets a second preset condition.

[0063] Optionally, the first candidate title determination submodule includes:

[0064] a correlation index calculation unit, configured to calculate the sum of similarities between a first paragraph title and other paragraph titles of the same paragraph to obtain a correlation index of the first paragraph title, wherein the first paragraph title is any one of at least two paragraph titles of the paragraph to which it belongs;

[0065] The first candidate title determining unit is used to determine that the first paragraph title is a first candidate title if the relevance index of the first paragraph title is greater than the relevance indexes of other paragraph titles of the paragraph to which it belongs.

[0066] Optionally, the second preset condition includes: the similarity between the second candidate title and the first candidate title is less than the similarity between the second paragraph title and the first candidate title, and the second paragraph title is any paragraph title that is not a candidate title among at least two paragraph titles of the paragraph to which the first candidate title belongs.

[0067] Optionally, the reference article acquisition module includes:

[0068] A data receiving submodule, for receiving a preset topic and screening conditions input by a user, wherein the screening conditions include at least one of article type, number of citations, publication time, article reading volume, and article like volume;

[0069] An article screening submodule, used to search for at least two reference articles that meet the preset topic according to the screening condition;

[0070] The paragraph determination submodule is used to determine the paragraphs of each reference article based on the pre-trained segmentation model.

[0071] Optionally, the reference article acquisition module further includes:

[0072] The segmentation module builds a submodule for building a segmentation model based on a bidirectional autoencoder model and a conditional random field;

[0073] A training corpus acquisition submodule is used to acquire segmented training corpus, wherein the segmented training corpus includes sample articles and segmentation points corresponding to the sample articles;

[0074] The segmentation model training submodule is used to train the segmentation model according to the segmentation training corpus until the segmentation model meets the convergence condition to obtain a trained segmentation model.

[0075] On the other hand, an embodiment of the present invention discloses a device for generating an article outline, the device comprising a memory and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by one or more processors, wherein the one or more programs include instructions for executing one or more of the article outline generation methods described above.

[0076] On the other hand, an embodiment of the present invention discloses a machine-readable medium having instructions stored thereon, which, when executed by one or more processors, enables the device to execute one or more of the article outline generation methods described above.

[0077] The embodiments of the present invention include the following advantages:

[0078] The embodiment of the present invention obtains at least two reference articles that meet the preset theme and determines the paragraphs of each reference article, and then further determines the candidate title corresponding to each paragraph of each reference article; clusters the candidate title corresponding to each paragraph of the at least two reference articles to obtain a global title connection graph corresponding to the preset theme; finally, generates the article outline corresponding to the preset theme according to the global title connection graph. When generating the article outline, the embodiment of the present invention obtains multiple reference articles that meet the preset theme, clusters the paragraph titles of each reference article, obtains a global title connection graph containing the paragraph titles of all reference articles, and finally generates the article outline based on the global title connection graph for user reference. Compared with the method of selecting an article from multiple articles in the prior art and generating the article outline based on the selected article, the article outline generated by the present invention can fully cover the information related to the preset theme in multiple articles that meet the preset theme, and improves the relevance of the generated article outline to the preset theme; and the present invention generates the article outline based on the paragraph titles of multiple reference articles, which enhances the credibility of the generated article outline and is conducive to improving the writing assistance effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0080] Figure 1 It is a flowchart of steps of an embodiment of a method for generating an article outline of the present invention;

[0081] Figure 2 It is a structural block diagram of an embodiment of an article outline generating device of the present invention;

[0082] Figure 3 is a block diagram of a device 800 for generating an article outline of the present invention;

[0083] Figure 4 It is a schematic diagram of the structure of the server in some embodiments of the present invention. DETAILED DESCRIPTION

[0084] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0085] Method Embodiment

[0086] Reference Figure 1 , shows a flowchart of a speech processing method embodiment of the present invention, the method may specifically include the following steps:

[0087] Step 101: Obtain at least two reference articles that meet a preset theme and determine the paragraphs of each reference article.

[0088] Step 102: Determine the candidate title corresponding to each paragraph of each reference article.

[0089] Step 103: clustering the candidate titles corresponding to each paragraph of the at least two reference articles to obtain a global title connection graph corresponding to the preset topic.

[0090] Step 104: Generate an article outline corresponding to the preset topic according to the global title connection diagram.

[0091] The article outline generation method provided in the embodiment of the present invention can be applied to a terminal device or a server. The terminal device may include but is not limited to: a smart terminal, a computer, a personal digital assistant (PDA), a tablet computer, an e-book reader, a laptop computer, a vehicle-mounted device, a smart TV, a wearable device, etc.

[0092] The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, cloud communications, network services, middleware services, content delivery networks (CDN), as well as big data and artificial intelligence platforms.

[0093] It should be noted that the preset theme can be a theme obtained by analyzing the writing requirements submitted by the user by the terminal device and / or the server. The writing requirements submitted by the user can be text information, and the terminal device and / or the server can determine the preset theme by performing semantic analysis on the text information; or, the writing requirements submitted by the user can also be voice information, and the terminal device and / or the server can first convert the voice information into text information through a built-in voice recognition model, and further perform semantic analysis on the converted text information to obtain the preset theme; or, the writing information submitted by the user can also be a picture, and the terminal device and / or the server can determine the content contained in the picture by performing feature analysis and text recognition on the picture, and then determine the preset theme based on the content contained in the picture, and so on.

[0094] The reference article may be any type of article that meets the preset theme. For example, the reference article may be a paper, press release, public account article, forum post, blog, web document, etc. that meets the preset theme. The embodiment of the present invention does not specifically limit the type of the reference article, as long as the reference article meets the preset theme.

[0095] When obtaining reference articles that meet the preset theme, the embodiment of the present invention can pre-build an article material database, and search for reference articles that meet the preset theme from the article material database. Among them, the article material database can use data services provided by third-party systems, such as various paper databases, literary databases, scientific and technological databases, etc. Specifically, free data can be collected from existing media platforms (including article retrieval platforms, mainstream self-media platforms, Weibo, forums, etc.) through web crawler technology, or authorized data can be collected in cooperation with a third party, or a third-party system can be directly used. Technical personnel can also regularly upload article materials to continuously update the article material database.

[0096] For each article in the article material database, the article content can be analyzed first to determine the article theme and generate corresponding theme tags. Then, when obtaining reference articles, the reference articles can be determined directly based on the similarity between the theme tags of each article and the preset theme.

[0097] After determining the reference articles that meet the preset theme, further determine the paragraphs of each article. Specifically, the paragraphs of each article can be determined based on a pre-trained segmentation model. The segmentation model can select any neural network model that is good at segmentation in the art, and the embodiment of the present invention does not make specific limitations on this.

[0098] Then, determine the candidate titles corresponding to each paragraph of each reference article. Specifically, the paragraph content can be analyzed to extract key entity words (such as named entity words such as place names, names of people, or other keywords) and subject words (such as sports, entertainment, emotions, etc.) in the paragraph, and then generate paragraph titles based on the extracted key entity words and subject words, and select a preset number of candidate titles from the generated multiple titles. Among them, the candidate titles can be selected from the multiple generated titles based on MMR (Maximal Marginal Relevance). Of course, other methods can also be used to select candidate titles.

[0099] After determining the candidate titles corresponding to each paragraph of each reference article, cluster processing is performed on the candidate titles corresponding to each paragraph of all reference articles to obtain a global title connection graph corresponding to the preset topic. The global title connection graph is composed of the candidate titles corresponding to each paragraph of all reference articles, and each node of the global title connection graph is generated according to the candidate titles, each candidate title corresponds to a node, and each corresponding node is connected according to the association relationship between each candidate title to obtain the global title connection graph.

[0100] Finally, an article outline corresponding to a preset topic is generated according to the global title connection graph. Specifically, an important subgraph, namely a target subgraph, in the global title connection graph is extracted, and an article outline is generated according to the titles corresponding to each node in the target subgraph.

[0101] Among them, the target subgraph contains the most relevant paragraph titles in multiple reference articles, which are in line with the writing ideas of multiple reference articles. Therefore, the article outline generated based on the target subgraph can cover the contents of multiple reference articles at the same time, thereby improving the relevance of the generated article outline to the preset theme; and, by generating an article outline based on the paragraph titles of multiple reference articles, the present invention enhances the credibility of the generated article outline, which is conducive to improving the writing assistance effect.

[0102] In an optional embodiment of the present invention, the step 101 of obtaining at least two reference articles that match a preset topic and determining a paragraph of each reference article includes:

[0103] Step S11, receiving a preset topic and screening conditions input by a user, wherein the screening conditions include at least one of article type, number of citations, publication time, article reading volume, and article like volume;

[0104] Step S12: searching for at least two reference articles that meet the preset theme according to the screening condition;

[0105] Step S13: Determine the paragraphs of each reference article based on the pre-trained segmentation model.

[0106] After the user inputs the proposed preset topic, he or she can further submit screening conditions to the terminal device and / or server according to actual needs and personal preferences to further clarify the requirements that the reference articles need to meet. The screening conditions may include at least one of the article type, number of citations, publication time, article reading volume, and article like volume. Among them, the article type may include the publication type of the article, such as papers (academic papers, dissertations, journal papers, etc.), newspaper and magazine articles, public account articles, forum posts, news reports, etc.; the article type may also be the writing type of newspaper articles, such as narrative, lyric, argumentative, expository, applied, etc.

[0107] In an embodiment of the present invention, the articles in the article material database can be analyzed in advance to determine the subject tags, type tags, time tags, popularity tags, etc. corresponding to each article, so as to facilitate searching for reference articles that meet the preset topics and screening conditions based on these tags.

[0108] After finding reference articles that meet the preset topics and screening conditions, it is necessary to further segment the reference articles to determine the paragraphs of each reference article. Specifically, the paragraphs of each article can be determined based on a pre-trained segmentation model. The segmentation model can be any neural network model that is good at segmentation in the field, and the embodiment of the present invention does not make specific limitations on this.

[0109] In order to be applicable to various types of articles and improve the accuracy of article segmentation results, a sequence labeling method can be used to build a segmentation model based on BERT (Bidirectional Encoder Representation from Transformers) and CRF (Conditional Random Field), and the constructed segmentation model can be trained. The reference articles can be segmented based on the trained segmentation model.

[0110] In an optional embodiment of the present invention, before determining the paragraphs of each reference article based on the pre-trained segmentation model in step S13, the method further includes:

[0111] Sub-step S131, constructing a segmentation model based on a bidirectional autoencoder model and a conditional random field;

[0112] Sub-step S132: obtaining segmentation training corpus, wherein the segmentation training corpus includes sample articles and segmentation points corresponding to the sample articles;

[0113] Sub-step S133: training the segmentation model according to the segmentation training corpus until the segmentation model meets a convergence condition to obtain a trained segmentation model.

[0114] Among them, BERT uses a masked language model to implement pre-trained deep bidirectional representation. The pre-trained BERT representation can be fine-tuned through an additional output layer, which is suitable for the construction of neural network models for a wide range of tasks without the need to make major modifications to the network architecture for specific tasks. Conditional Random Field (CRF) is a sequence labeling algorithm, which can also be regarded as a discriminant model that can predict the input sequence based on the input sequence. The segmentation model built based on BERT and CRF can be trained with a small amount of labeled data and quickly migrated to new application scenarios. It can be applied to the segmentation processing of various articles.

[0115] Of course, in the embodiment of the present invention, other neural network models that are good at article segmentation may also be selected, and the embodiment of the present invention does not make specific limitations on this.

[0116] In an optional embodiment of the present invention, the candidate titles in step 102 include a first candidate title and a second candidate title, and determining the candidate title corresponding to each paragraph of each reference article includes:

[0117] Step S21, generating at least two paragraph titles for each paragraph of each reference article;

[0118] Step S22, determining a first candidate title from each paragraph title of each paragraph, wherein the similarity between the first candidate title and other paragraph titles of the paragraph to which it belongs meets a first preset condition;

[0119] Step S23: Determine a second candidate title whose similarity to the first candidate title meets a second preset condition.

[0120] When generating a paragraph title, if the paragraph title is generated only based on the content of the paragraph itself, there may be a deviation between the obtained paragraph title and the actual content of the paragraph. Therefore, in an embodiment of the present invention, the full-text keywords of the reference article can be determined by the full-text content of the reference article, and then the paragraph title is generated based on the full-text keywords and the content of the paragraph itself to improve the accuracy of the paragraph title.

[0121] There are often multiple full-text keywords in the reference article, and there may also be multiple title generation models for generating paragraph titles, so multiple paragraph titles can be obtained for the same paragraph. There are often some commonalities or similarities between these obtained paragraph titles. Therefore, in order to avoid data redundancy and improve processing efficiency, the multiple paragraph titles obtained can be screened, and some of the titles can be selected as candidate titles, and an article outline can be generated based on the selected candidate titles. The number of selected candidate titles can be set according to actual needs. For example, 3 titles are selected as candidate titles. The embodiment of the present invention does not specifically limit the number of candidate titles.

[0122] In order to ensure the credibility of the final generated article outline, the selected candidate titles must be able to represent the paragraph content. Therefore, in the embodiment of the present invention, the candidate titles are determined based on the similarity between the paragraph titles. The first candidate title is the first candidate title selected. After the first candidate title is determined, other candidate titles, i.e., the second candidate titles, are further determined based on the similarity between the paragraph title and the first candidate title.

[0123] Specifically, candidate titles can be extracted based on MMR (Maximal Marginal Relevance). Then, the first preset condition is that the sum of the similarities between the first candidate title and other paragraph titles is the highest, and the second preset condition is that the similarity between the second candidate title and the first candidate title is the lowest. In this way, while ensuring that the selected candidate title can represent the paragraph content, the diversity of the candidate titles can be improved, thereby avoiding data redundancy while covering as much information as possible, which is conducive to improving the credibility of the final generated article outline.

[0124] Of course, other methods may also be used to select candidate titles. The first preset condition and the second preset condition may be determined according to a specific method for selecting candidate titles, and the embodiment of the present invention does not specifically limit this.

[0125] Taking the selection of candidate titles based on the MMR algorithm as an example, in an optional embodiment of the present invention, step S22, determining the first candidate title from each paragraph title of each paragraph, includes:

[0126] Sub-step S221, calculating the sum of similarities between the first paragraph title and other paragraph titles of the same paragraph to obtain a correlation index of the first paragraph title, where the first paragraph title is any one of at least two paragraph titles of the paragraph to which it belongs;

[0127] Sub-step S222: If the relevance index of the first paragraph title is greater than the relevance indexes of other paragraph titles of the paragraph to which it belongs, then the first paragraph title is determined to be the first candidate title.

[0128] When determining the first candidate title, first calculate the sum of the similarities between each paragraph title and other paragraph titles of the same paragraph, and obtain the correlation index of each paragraph title. Then, compare the correlation indexes of each paragraph title, and determine the paragraph title with the largest correlation index as the first candidate title. In this way, it can be determined that the selected first candidate title has the highest correlation with each generated paragraph title, that is, it contains the common points between each paragraph title, so that it can represent the paragraph content to the maximum extent.

[0129] In an optional embodiment of the present invention, the second preset condition includes: the similarity between the second candidate title and the first candidate title is less than the similarity between the second paragraph title and the first candidate title, and the second paragraph title is any paragraph title that is not a candidate title among at least two paragraph titles of the paragraph to which the first candidate title belongs.

[0130] After selecting the first candidate title that can represent the content of the paragraph, in order to further supplement the information of the first candidate title and improve the accuracy of the selected candidate title, the similarity between each paragraph title and the first candidate title can be continued to be calculated, and the paragraph title with the lowest similarity to the first candidate title can be selected as the second candidate title, thereby improving the diversity of the candidate titles.

[0131] In an optional embodiment of the present invention, the step 103 of clustering the candidate titles corresponding to each paragraph of the at least two reference articles to obtain a global title connection graph corresponding to the preset topic includes:

[0132] Step S31, merging the candidate titles corresponding to each paragraph to obtain the target title corresponding to each paragraph;

[0133] Step S32: connecting the target titles corresponding to the paragraphs of the same reference article in sequence according to the order of the paragraphs to obtain at least two internal title connection graphs, each of which corresponds to a reference article;

[0134] Step S33: clustering the target titles in the at least two internal title connection graphs according to the semantics of the target titles to obtain a global title connection graph corresponding to the preset theme.

[0135] When clustering candidate titles of each paragraph of each reference article, in order to reduce the amount of data processing and improve data processing efficiency, the candidate titles corresponding to each paragraph can be merged first, and multiple candidate titles can be merged into a target title. The merged target title contains the information of each candidate title of the paragraph. Specifically, the common features of each candidate title can be extracted, and the extracted common features and the distinguishing features of each candidate title can be combined to generate a target title that conforms to grammatical semantics.

[0136] Then, the target titles belonging to the same reference article are connected in sequence according to the order of the corresponding paragraphs in the reference article to obtain an internal title connection diagram corresponding to the reference article, and each reference article corresponds to an internal title connection diagram. Specifically, when generating the target titles, the paragraph position information corresponding to each target title and the reference article information to which it belongs can be recorded, so that based on the paragraph position information corresponding to each target title and the reference article information to which it belongs, the target titles corresponding to each paragraph of the same reference article are connected in sequence according to the order of the paragraphs.

[0137] It should be noted that the internal title connection diagram reflects the internal connection relationship between each target title of the reference article. The global title connection diagram not only includes the internal connection relationship between each target title belonging to the same reference article, but also includes the external connection relationship between each target title belonging to different reference articles. Therefore, after generating the internal title connection diagram, it is necessary to further perform clustering processing on all the target titles in each generated internal title connection diagram, and determine the external connection relationship between the target titles according to the clusters to which each target title of different reference articles belongs.

[0138] In an optional embodiment of the present invention, the step S33 of clustering the target titles in the at least two internal title connection graphs according to the semantics of the target title to obtain a global title connection graph corresponding to the preset topic includes:

[0139] Sub-step S331, clustering the target titles belonging to different reference articles according to the semantics of the target title to obtain the cluster to which each target title belongs;

[0140] Sub-step S332: Connect the target titles belonging to the same cluster in the at least two internal title connection graphs to obtain a global title connection graph corresponding to the preset theme.

[0141] In the embodiment of the present invention, the target titles belonging to different reference articles are clustered according to the semantics of the target titles, the clusters to which the target titles belong are determined, and then the target titles belonging to the same cluster in the internal title connection graphs are connected to obtain a global title connection graph. The length of the external connection line between the target titles in the same cluster in the global title connection graph can be determined according to the similarity between the target titles in the cluster. For example, the higher the similarity between the two target titles, the shorter the external connection line between them, indicating a closer relationship, and so on.

[0142] In an optional embodiment of the present invention, the step S332 of connecting the target titles belonging to the same cluster in the at least two internal title connection graphs to obtain a global title connection graph corresponding to the preset theme includes:

[0143] Step P11, determining whether there are at least two identical target titles in the same cluster of the at least two internal title connection graphs;

[0144] Step P12: if there are at least two identical target titles in the same cluster, the at least two identical target titles are merged to obtain a merged target title, and the target title in the cluster to which it belongs is updated according to the merged target title;

[0145] Step P13: Connect the updated target titles in the same cluster of the at least two internal title connection graphs to obtain a global title connection graph corresponding to the preset theme.

[0146] It should be noted that, in actual applications, there are cases where the target titles of different reference articles are the same. For example, for reference article A, it has corresponding target titles a1, a2, and a3, and for reference article B, it has corresponding target titles b1, b2, b3, and b4. Among them, the target title a2 of reference article A and the target title b3 of reference article A belong to the same cluster, and the target title a2 is the same as the target title b3. Then, in an embodiment of the present invention, in order to simplify the global title connection diagram and avoid data redundancy, the same target title a2 and target title b3 can be merged to generate a new target title. For example, the newly generated target title can be recorded as a2b3, and then the target titles of the same cluster are connected to obtain a global title connection diagram.

[0147] In an optional embodiment of the present invention, the step 104 of generating an article outline corresponding to the preset topic according to the global title connection diagram includes:

[0148] Step S41, extracting a target subgraph from the global title connection graph based on a community discovery algorithm;

[0149] Step S42: Generate an article outline corresponding to the preset topic according to the target subgraph.

[0150] Among them, the target subgraph contains the most relevant paragraph titles in multiple reference articles, which conforms to the writing ideas of multiple reference articles, so that the article outline generated based on the target subgraph can cover the contents of multiple reference articles at the same time, and improves the relevance of the generated article outline to the preset topic. The target subgraph in the global title connection graph can be extracted based on the community discovery algorithm. Specifically, by measuring the similarity between each target title inside the cluster in each subgraph contained in the global title connection graph and the similarity outside the cluster, a subgraph with high similarity of target titles inside the cluster and low similarity of target titles outside the cluster is selected as the target subgraph.

[0151] It should be noted that there are semantically similar target titles in the target subgraph, so multiple article outlines can be generated according to different branches in the target subgraph. In practical applications, the generated multiple article outlines can be recommended to the user at the same time for the user to choose.

[0152] In summary, when generating an article outline, the embodiment of the present invention obtains multiple reference articles that meet the preset theme, and clusters the paragraph titles of each reference article to obtain a global title connection graph containing the paragraph titles of all reference articles. Finally, an article outline is generated based on the global title connection graph for user reference. Compared with the method in the prior art of screening out an article from multiple articles and generating an article outline based on the screened article, the article outline generated by the present invention can fully cover the information related to the preset theme in multiple articles that meet the preset theme, thereby improving the relevance of the generated article outline to the preset theme; and, by generating an article outline based on the paragraph titles of multiple reference articles, the present invention enhances the credibility of the generated article outline, which is conducive to improving the writing assistance effect.

[0153] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0154] Device Embodiment

[0155] Reference Figure 2 , shows a structural block diagram of an embodiment of an article outline generation device of the present invention, the device may include:

[0156] A reference article acquisition module 201 is used to acquire at least two reference articles that meet a preset theme and determine a paragraph of each reference article;

[0157] A candidate title determination module 202 is used to determine a candidate title corresponding to each paragraph of each reference article;

[0158] A title connection graph generating module 203 is used to perform clustering processing on the candidate titles corresponding to each paragraph of the at least two reference articles to obtain a global title connection graph corresponding to the preset topic;

[0159] The article outline generating module 204 is used to generate the article outline corresponding to the preset theme according to the global title connection diagram.

[0160] Optionally, the title connection diagram generating module includes:

[0161] The title merging submodule is used to merge the candidate titles corresponding to each paragraph to obtain the target title corresponding to each paragraph;

[0162] An internal title connection graph generation submodule is used to sequentially connect the target titles corresponding to the paragraphs of the same reference article according to the order of the paragraphs to obtain at least two internal title connection graphs, each of which corresponds to a reference article;

[0163] The global title connection graph generating submodule is used to perform clustering processing on each target title in the at least two internal title connection graphs according to the semantics of the target title to obtain a global title connection graph corresponding to the preset theme.

[0164] Optionally, the global title connection graph generation submodule includes:

[0165] A clustering processing unit, used for clustering target titles belonging to different reference articles according to the semantics of the target titles to obtain the cluster to which each target title belongs;

[0166] The global title connection graph generating unit is used to connect the target titles belonging to the same cluster in the at least two internal title connection graphs to obtain the global title connection graph corresponding to the preset theme.

[0167] Optionally, the global title connection graph generating unit includes:

[0168] a target title determination subunit, used to determine whether there are at least two identical target titles in the same cluster of the at least two internal title connection graphs;

[0169] A target title merging subunit is used for merging at least two identical target titles to obtain a merged target title if there are at least two identical target titles in the same cluster, and updating the target title in the cluster to which it belongs according to the merged target title;

[0170] The target title connection subunit is used to connect the updated target titles in the same cluster of the at least two internal title connection graphs to obtain a global title connection graph corresponding to the preset theme.

[0171] Optionally, the article outline generating module includes:

[0172] A target subgraph extraction submodule, used for extracting a target subgraph from the global title connection graph based on a community discovery algorithm;

[0173] The article outline generation submodule is used to generate an article outline corresponding to the preset theme according to the target subgraph.

[0174] Optionally, the candidate title includes a first candidate title and a second candidate title, and the candidate title determination module includes:

[0175] A paragraph title generation submodule, used for generating at least two paragraph titles for each paragraph of each reference article;

[0176] A first candidate title determination submodule, used to determine a first candidate title from each paragraph title of each paragraph, wherein the similarity between the first candidate title and other paragraph titles of the paragraph to which it belongs meets a first preset condition;

[0177] The second candidate title determination submodule is used to determine a second candidate title whose similarity with the first candidate title meets a second preset condition.

[0178] Optionally, the first candidate title determination submodule includes:

[0179] a correlation index calculation unit, configured to calculate the sum of similarities between a first paragraph title and other paragraph titles of the same paragraph to obtain a correlation index of the first paragraph title, wherein the first paragraph title is any one of at least two paragraph titles of the paragraph to which it belongs;

[0180] The first candidate title determining unit is used to determine that the first paragraph title is a first candidate title if the relevance index of the first paragraph title is greater than the relevance indexes of other paragraph titles of the paragraph to which it belongs.

[0181] Optionally, the second preset condition includes: the similarity between the second candidate title and the first candidate title is less than the similarity between the second paragraph title and the first candidate title, and the second paragraph title is any paragraph title that is not a candidate title among at least two paragraph titles of the paragraph to which the first candidate title belongs.

[0182] Optionally, the reference article acquisition module includes:

[0183] A data receiving submodule, for receiving a preset topic and screening conditions input by a user, wherein the screening conditions include at least one of article type, number of citations, publication time, article reading volume, and article like volume;

[0184] An article screening submodule, used to search for at least two reference articles that meet the preset topic according to the screening condition;

[0185] The paragraph determination submodule is used to determine the paragraphs of each reference article based on the pre-trained segmentation model.

[0186] Optionally, the reference article acquisition module further includes:

[0187] The segmentation module builds a submodule for building a segmentation model based on a bidirectional autoencoder model and a conditional random field;

[0188] A training corpus acquisition submodule is used to acquire segmented training corpus, wherein the segmented training corpus includes sample articles and segmentation points corresponding to the sample articles;

[0189] The segmentation model training submodule is used to train the segmentation model according to the segmentation training corpus until the segmentation model meets the convergence condition to obtain a trained segmentation model.

[0190] In summary, when generating an article outline, the embodiment of the present invention obtains multiple reference articles that meet the preset theme, and clusters the paragraph titles of each reference article to obtain a global title connection graph containing the paragraph titles of all reference articles. Finally, an article outline is generated based on the global title connection graph for user reference. Compared with the method in the prior art of screening out an article from multiple articles and generating an article outline based on the screened article, the article outline generated by the present invention can fully cover the information related to the preset theme in multiple articles that meet the preset theme, thereby improving the relevance of the generated article outline to the preset theme; and, by generating an article outline based on the paragraph titles of multiple reference articles, the present invention enhances the credibility of the generated article outline, which is conducive to improving the writing assistance effect.

[0191] As for the device embodiment, since it is 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.

[0192] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0193] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0194] An embodiment of the present invention provides a device for generating an article outline, the device comprising a memory and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by one or more processors. The one or more programs include instructions for performing the following operations:

[0195] A11. Obtain at least two reference articles that match the preset topic and determine the paragraphs of each reference article;

[0196] A12. Determine the candidate title corresponding to each paragraph of each reference article;

[0197] A13, clustering the candidate titles corresponding to each paragraph of the at least two reference articles to obtain a global title connection graph corresponding to the preset topic;

[0198] A14. Generate an article outline corresponding to the preset topic according to the global title connection diagram.

[0199] Optionally, clustering the candidate titles corresponding to each paragraph of the at least two reference articles to obtain a global title connection graph corresponding to the preset topic includes:

[0200] Merge the candidate titles corresponding to each paragraph to obtain the target title corresponding to each paragraph;

[0201] Connecting the target titles corresponding to the paragraphs of the same reference article in sequence according to the order of the paragraphs to obtain at least two internal title connection graphs, each of which corresponds to a reference article;

[0202] Clustering is performed on each target title in the at least two internal title connection graphs according to the semantics of the target title to obtain a global title connection graph corresponding to the preset theme.

[0203] Optionally, clustering the target titles in the at least two internal title connection graphs according to the semantics of the target titles to obtain a global title connection graph corresponding to the preset topic includes:

[0204] Clustering the target titles belonging to different reference articles according to the semantics of the target title to obtain the cluster to which each target title belongs;

[0205] The target titles belonging to the same cluster in the at least two internal title connection graphs are connected to obtain a global title connection graph corresponding to the preset theme.

[0206] Optionally, the step of connecting the target titles belonging to the same cluster in the at least two internal title connection graphs to obtain a global title connection graph corresponding to the preset theme includes:

[0207] Determining whether there are at least two identical target titles in the same cluster of the at least two internal title connection graphs;

[0208] If there are at least two identical target titles in the same cluster, the at least two identical target titles are merged to obtain a merged target title, and the target title in the cluster to which it belongs is updated according to the merged target title;

[0209] The updated target titles in the same cluster of the at least two internal title connection graphs are connected to obtain a global title connection graph corresponding to the preset theme.

[0210] Optionally, generating an article outline corresponding to the preset topic according to the global title connection diagram includes:

[0211] Extracting a target subgraph in the global title connection graph based on a community discovery algorithm;

[0212] Generate an article outline corresponding to the preset topic according to the target subgraph.

[0213] Optionally, the candidate titles include a first candidate title and a second candidate title, and determining the candidate title corresponding to each paragraph of each reference article includes:

[0214] For each paragraph of each reference article, generate at least two paragraph headings;

[0215] Determine a first candidate title from each paragraph title of each paragraph, wherein the first candidate title is a first candidate title whose similarity with other paragraph titles of the paragraph to which it belongs satisfies a first preset condition;

[0216] A second candidate title is determined, the similarity of which with the first candidate title satisfies a second preset condition.

[0217] Optionally, determining the first candidate title from each paragraph title of each paragraph includes:

[0218] Calculating the sum of similarities between the first paragraph title and other paragraph titles of the same paragraph to obtain a correlation index of the first paragraph title, where the first paragraph title is any one of at least two paragraph titles of the paragraph to which it belongs;

[0219] If the relevance index of the first paragraph title is greater than the relevance indexes of other paragraph titles of the paragraph to which it belongs, the first paragraph title is determined to be the first candidate title.

[0220] Optionally, the second preset condition includes: the similarity between the second candidate title and the first candidate title is less than the similarity between the second paragraph title and the first candidate title, and the second paragraph title is any paragraph title that is not a candidate title among at least two paragraph titles of the paragraph to which the first candidate title belongs.

[0221] Optionally, obtaining at least two reference articles that match a preset topic and determining a paragraph of each reference article includes:

[0222] Receive a preset topic and screening conditions input by a user, wherein the screening conditions include at least one of article type, number of citations, publication time, article reading volume, and article like volume;

[0223] Searching for at least two reference articles that meet the preset topic according to the screening condition;

[0224] The paragraphs of each reference article are determined based on a pre-trained segmentation model.

[0225] Optionally, the device is further configured to execute, by one or more processors, the one or more programs including instructions for performing the following operations:

[0226] Construct a segmentation model based on a bidirectional autoencoder model and conditional random fields;

[0227] Acquire segmented training corpus, wherein the segmented training corpus includes sample articles and segmentation points corresponding to the sample articles;

[0228] The segmentation model is trained according to the segmentation training corpus until the segmentation model meets a convergence condition to obtain a trained segmentation model.

[0229] Figure 3 8 is a block diagram of an apparatus 800 for generating an article outline according to an exemplary embodiment. For example, the apparatus 800 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0230] Reference Figure 3, the device 800 may include one or more of the following components: a processing component 802 , a memory 804 , a power component 806 , a multimedia component 808 , an audio component 810 , an input / output (I / O) interface 812 , a sensor component 814 , and a communication component 816 .

[0231] The processing component 802 generally controls the overall operation of the device 800, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the above-mentioned method. In addition, the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components. For example, the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

[0232] The memory 804 is configured to store various types of data to support operations on the device 800. Examples of such data include instructions for any application or method operating on the device 800, contact data, phone book data, messages, pictures, videos, etc. The memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0233] The power supply component 806 provides power to the various components of the device 800. The power supply component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the device 800.

[0234] The multimedia component 808 includes a screen that provides an output interface between the device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. When the device 800 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0235] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC), and when the device 800 is in an operating mode, such as a call mode, a recording mode, and a voice information processing mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 804 or sent via the communication component 816. In some embodiments, the audio component 810 also includes a speaker for outputting audio signals.

[0236] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: home button, volume button, start button, and lock button.

[0237] The sensor assembly 814 includes one or more sensors for providing various aspects of status assessment for the device 800. For example, the sensor assembly 814 can detect the open / closed state of the device 800, the relative positioning of components, such as the display and keypad of the device 800, the sensor assembly 814 can also detect the position change of the voice processing device 800 or a component of the device 800, the presence or absence of user contact with the device 800, the orientation or acceleration / deceleration of the device 800, and the temperature change of the device 800. The sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. The sensor assembly 814 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an accelerometer, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0238] The communication component 816 is configured to facilitate wired or wireless communication between the device 800 and other devices. The device 800 can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency information processing (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0239] In an exemplary embodiment, the apparatus 800 may be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0240] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 804 including instructions, and the instructions can be executed by the processor 820 of the device 800 to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0241] Figure 4 1 is a schematic diagram of the structure of the server in some embodiments of the present invention. The server 1900 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 1922 (for example, one or more processors) and memory 1932, and one or more storage media 1930 (for example, one or more mass storage devices) storing application programs 1942 or data 1944. Among them, the memory 1932 and the storage medium 1930 can be short-term storage or permanent storage. The program stored in the storage medium 1930 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the server. Furthermore, the central processing unit 1922 can be configured to communicate with the storage medium 1930 and execute a series of instruction operations in the storage medium 1930 on the server 1900.

[0242] The server 1900 may also include one or more power supplies 1926, one or more wired or wireless network interfaces 1950, one or more input and output interfaces 1958, one or more keyboards 1956, and / or, one or more operating systems 1941, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

[0243] A non-transitory computer-readable storage medium, when the instructions in the storage medium are executed by a processor of a device (server or terminal), enables the device to perform Figure 1 The article outline generation method shown.

[0244] A non-temporary computer-readable storage medium, when the instructions in the storage medium are executed by a processor of a device (server or terminal), enables the device to execute an article outline generation method, the method comprising: obtaining at least two reference articles that conform to a preset theme and determining the paragraphs of each reference article; determining a candidate title corresponding to each paragraph of each reference article; clustering the candidate titles corresponding to each paragraph of the at least two reference articles to obtain a global title connection graph corresponding to the preset theme; and generating an article outline corresponding to the preset theme according to the global title connection graph.

[0245] Those skilled in the art will readily appreciate other embodiments of the invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed in this disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the following claims.

[0246] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

[0247] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

[0248] The above is a detailed introduction to an article outline generating method, an article outline generating device and a device for generating an article outline provided by the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for generating an article outline, characterized in that: The method comprises: Obtain at least two reference articles that match the pre-set topic and identify the paragraphs of each reference article; Determine a candidate title corresponding to each paragraph of each reference article; the candidate title includes a first candidate title and a second candidate title, and the determining of the candidate title corresponding to each paragraph of each reference article includes: For each paragraph of each reference article, generate at least two paragraph headings; Determining a first candidate title from each paragraph title of each paragraph, wherein the similarity between the first candidate title and other paragraph titles of the paragraph to which it belongs meets a first preset condition, including: calculating the sum of similarities between the first paragraph title and other paragraph titles of the same paragraph to obtain a correlation index of the first paragraph title, wherein the first paragraph title is any one of at least two paragraph titles of the paragraph to which it belongs; if the correlation index of the first paragraph title is greater than the correlation indexes of other paragraph titles of the paragraph to which it belongs, determining the first paragraph title as the first candidate title; Determine a second candidate title whose similarity to the first candidate title satisfies a second preset condition; the second preset condition includes: the similarity between the second candidate title and the first candidate title is less than the similarity between a second paragraph title and the first candidate title, and the second paragraph title is any paragraph title that is not a candidate title among at least two paragraph titles of the paragraph to which the first candidate title belongs; The candidate titles corresponding to each paragraph of the at least two reference articles are clustered to obtain a global title connection graph corresponding to the preset theme, including: merging the candidate titles corresponding to each paragraph to obtain a target title corresponding to each paragraph; sequentially connecting the target titles corresponding to the paragraphs of the same reference article in sequence to obtain at least two internal title connection graphs, each of which corresponds to a reference article; clustering the target titles in the at least two internal title connection graphs according to the semantics of the target title to obtain a global title connection graph corresponding to the preset theme; An article outline corresponding to the preset topic is generated according to the global title connection diagram.

2. The method according to claim 1, characterized in that: The step of clustering the target titles in the at least two internal title connection graphs according to the semantics of the target title to obtain a global title connection graph corresponding to the preset topic includes: Clustering the target titles belonging to different reference articles according to the semantics of the target title to obtain the cluster to which each target title belongs; The target titles belonging to the same cluster in the at least two internal title connection graphs are connected to obtain a global title connection graph corresponding to the preset theme.

3. The method according to claim 2, characterized in that The step of connecting the target titles belonging to the same cluster in the at least two internal title connection graphs to obtain a global title connection graph corresponding to the preset theme includes: Determining whether there are at least two identical target titles in the same cluster of the at least two internal title connection graphs; If there are at least two identical target titles in the same cluster, the at least two identical target titles are merged to obtain a merged target title, and the target title in the cluster to which it belongs is updated according to the merged target title; The updated target titles in the same cluster of the at least two internal title connection graphs are connected to obtain a global title connection graph corresponding to the preset theme.

4. The method according to claim 1, characterized in that Generating the article outline corresponding to the preset topic according to the global title connection diagram includes: Extracting a target subgraph in the global title connection graph based on a community discovery algorithm; the target subgraph in the global title connection graph is an important subgraph in the global title connection graph, including the most relevant paragraph titles in multiple reference articles, and conforming to the writing ideas of multiple reference articles; Generate an article outline corresponding to the preset topic according to the target subgraph.

5. The method according to claim 1, characterized in that The step of obtaining at least two reference articles that meet the preset topic and determining the paragraphs of each reference article includes: Receive a preset topic and screening conditions input by a user, wherein the screening conditions include at least one of article type, number of citations, publication time, article reading volume, and article like volume; Searching for at least two reference articles that meet the preset topic according to the screening condition; The paragraphs of each reference article are determined based on a pre-trained segmentation model.

6. The method according to claim 5, characterized in that Before determining the paragraphs of each reference article based on the pre-trained segmentation model, the method further includes: Construct a segmentation model based on a bidirectional autoencoder model and conditional random fields; Acquire segmented training corpus, wherein the segmented training corpus includes sample articles and segmentation points corresponding to the sample articles; The segmentation model is trained according to the segmentation training corpus until the segmentation model meets a convergence condition to obtain a trained segmentation model.

7. An article outline generating device, characterized in that: The device comprises: A reference article acquisition module is used to acquire at least two reference articles that meet the preset theme and determine the paragraphs of each reference article; A candidate title determination module is used to determine a candidate title corresponding to each paragraph of each reference article; the candidate title includes a first candidate title and a second candidate title, and the candidate title determination module includes: A paragraph title generation submodule, used for generating at least two paragraph titles for each paragraph of each reference article; A first candidate title determination submodule is used to determine a first candidate title from each paragraph title of each paragraph, wherein the similarity between the first candidate title and other paragraph titles of the paragraph to which it belongs meets a first preset condition, and includes: a correlation index calculation unit, used to calculate the sum of similarities between the first paragraph title and other paragraph titles of the same paragraph to obtain a correlation index of the first paragraph title, wherein the first paragraph title is any one of at least two paragraph titles of the paragraph to which it belongs; and a first candidate title determination unit, used to determine that the first paragraph title is the first candidate title if the correlation index of the first paragraph title is greater than the correlation indexes of other paragraph titles of the paragraph to which it belongs; A second candidate title determination submodule is used to determine a second candidate title whose similarity to the first candidate title satisfies a second preset condition; the second preset condition includes: the similarity between the second candidate title and the first candidate title is less than the similarity between the second paragraph title and the first candidate title, and the second paragraph title is any paragraph title that is not a candidate title among at least two paragraph titles of the paragraph to which the first candidate title belongs; The title connection graph generation module is used to perform clustering processing on the candidate titles corresponding to each paragraph of the at least two reference articles to obtain a global title connection graph corresponding to the preset topic, including: The title merging submodule is used to merge the candidate titles corresponding to each paragraph to obtain the target title corresponding to each paragraph; An internal title connection graph generation submodule is used to sequentially connect the target titles corresponding to the paragraphs of the same reference article according to the order of the paragraphs to obtain at least two internal title connection graphs, each of which corresponds to a reference article; A global title connection graph generating submodule, used for clustering each target title in the at least two internal title connection graphs according to the semantics of the target title, to obtain a global title connection graph corresponding to the preset topic; An article outline generation module is used to generate an article outline corresponding to the preset theme according to the global title connection diagram.

8. The device according to claim 7, characterized in that The global title connection graph generation submodule includes: A clustering processing unit, used for clustering target titles belonging to different reference articles according to the semantics of the target titles to obtain the cluster to which each target title belongs; The global title connection graph generating unit is used to connect the target titles belonging to the same cluster in the at least two internal title connection graphs to obtain the global title connection graph corresponding to the preset theme.

9. A device for generating an article outline, characterized in that: The device includes a memory and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by one or more processors. The one or more programs include instructions for executing the article outline generation method as described in any one of claims 1 to 6.

10. A machine-readable medium having instructions stored thereon, which, when executed by one or more processors, causes a device to execute the article outline generating method as claimed in any one of claims 1 to 6.

Images