Article content scoring method and device

Abstract

The present disclosure provides an article content scoring method and device. The article content scoring device collects articles published by users in a scene with a preset theme; uses a first deep learning model to score the content and scene theme of the article to obtain a first scoring result; extracts the text statistical features of the article; extracts the associated features of the author of the article; uses a second deep learning model to score the text statistical features of the article and the associated features of the author to obtain a second scoring result; and fuses the first scoring result and the second scoring result to obtain the scoring result of the article. In the scoring process, the present disclosure can more accurately reflect the quality of the article by combining the author's associated information and the degree of matching between the article theme and the theme of the publishing scene.

Description

Technical Field

[0001] The present disclosure relates to the field of information processing, and in particular to a method and device for scoring article content. Background Art

[0002] Content marketing has been valued by many businesses due to its advantages of good communication and low customer acquisition costs, which has led to a huge explosion of online articles. In e-commerce WeChat shopping circles such as JD.com, there are multiple circles according to different themes, such as mommy circle, foodie circle, music circle, photography circle, etc., each circle is shared by fashion experts, businesses, celebrities, users, etc. to share their shopping experiences. Usually, articles are rated and high-quality articles are selected and pinned to the top to achieve better communication effects. Summary of the invention

[0003] The inventors found through research that in the related art, articles are scored only based on their vocabulary features, literary features, concept features and paragraph features. Since the influence of the author on the dissemination of the article and whether the theme of the article matches the theme of the published scene are not considered, the article scoring is greatly biased.

[0004] To this end, the present disclosure provides a scheme for scoring articles according to the authors and topics of the articles, so as to more accurately reflect the quality of the articles.

[0005] According to a first aspect of an embodiment of the present disclosure, a method for scoring article content is provided, comprising: collecting articles published by users in a scene with a preset theme; scoring the content of the articles and the theme of the scene using a first deep learning model to obtain a first scoring result; extracting text statistical features of the articles; extracting associated features of the author of the articles; scoring the text statistical features of the articles and the associated features of the author using a second deep learning model to obtain a second scoring result; and fusing the first scoring result and the second scoring result to obtain a scoring result for the article.

[0006] In some embodiments, using the first deep learning model to score the content of the article and the theme of the scene includes: performing word segmentation on the text of the article to obtain corresponding word vectors; analyzing the theme of the article to obtain the theme vector of the article; and using the first deep learning model to score the word vectors, the theme vectors, and the theme of the scene to obtain the first scoring result.

[0007] In some embodiments, analyzing the topic of the article includes: analyzing the topic of the article using a latent Dirichlet distribution model to obtain probability values ​​of the article on each topic; and generating a topic vector of the article based on the probability values ​​of the article on each topic.

[0008] In some embodiments, the first deep learning model includes multiple sub-models, and the types of each sub-model are different; using the first deep learning model to score the word vector, the topic vector, and the theme of the scene includes: inputting the word vector, the topic vector, and the theme of the scene into multiple sub-models for scoring processing, so that each sub-model outputs a corresponding scoring result; and fusing the scoring results output by each sub-model to obtain the first scoring result.

[0009] In some embodiments, the multiple sub-models include at least one of an Inception model, an RCNN model, or a Fast Text model.

[0010] In some embodiments, the text statistical features of the article include: at least one of the number of characters, the number of words, the number of paragraphs, or the number of pictures in the article.

[0011] In some embodiments, the associated features of the author include attribute information of the author, wherein the attribute information of the author includes at least one of the author's level, number of fans, number of followers, the difference between the number of fans and the number of followers, or the ratio of the number of fans to the number of followers.

[0012] In some embodiments, the associated features of the author also include evaluation information of the author's published articles, wherein the evaluation information of the author's published articles includes at least one of the total number of articles published by the author, the total number of articles with high-quality tags added in the published articles, the proportion of the total number of articles with high-quality tags added in the published articles, the average number of views of published articles, the number of single visitors, the average stay time, the average number of likes, the average number of comments or the average number of reposts.

[0013] In some embodiments, the evaluation information of the author's published articles further includes an impact evaluation index, wherein the impact evaluation index is associated with the number of likes, the number of comments, and the number of single visitors.

[0014] In some embodiments, the author's association feature further includes the author's trend feature, wherein the trend feature includes the amount of change in at least one of the author's attribute information or the published article evaluation information within the past T1 days relative to the past T2 days, wherein T1 <T2。

[0015] In some embodiments, the second deep learning model comprises an xgboost model or a lightgbm model.

[0016] According to a second aspect of an embodiment of the present disclosure, there is provided an article content scoring device, comprising: a collection module, configured to collect articles published by users in a scene with a preset theme; a first scoring processing module, configured to use a first deep learning model to score the content of the article and the theme of the scene to obtain a first scoring result; a second scoring processing module, configured to extract text statistical features of the article, extract associated features of the author of the article, and use a second deep learning model to score the text statistical features of the article and the associated features of the author to obtain a second scoring result; a scoring module, configured to fuse the first scoring result and the second scoring result to obtain a scoring result for the article.

[0017] According to a third aspect of an embodiment of the present disclosure, there is provided an article content scoring device, comprising: a memory configured to store instructions; a processor coupled to the memory, the processor being configured to execute a method as described in any of the above embodiments based on the instructions stored in the memory.

[0018] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the method involved in any of the above embodiments is implemented.

[0019] Other features and advantages of the present disclosure will become apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present disclosure 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 disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0021] Figure 1 A schematic diagram of a flow chart of a method for scoring article content according to an embodiment of the present disclosure;

[0022] Figure 2 A schematic diagram of a process for scoring the content and scene themes of an article according to an embodiment of the present disclosure;

[0023] Figure 3 This is a schematic diagram of the structure of an article content scoring device according to an embodiment of the present disclosure;

[0024] Figure 4 It is a structural schematic diagram of an article content scoring device according to another embodiment of the present disclosure. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present disclosure and its application or use. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0026] Unless specifically stated otherwise, the relative arrangement of components and steps, the numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present disclosure.

[0027] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0028] Technologies, methods, and apparatus known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, such technologies, methods, and apparatus should be considered part of the authorization specification.

[0029] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0030] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0031] Figure 1 The flowchart of the article content scoring method according to one embodiment of the present disclosure is shown in FIG.

[0032] In step 101, articles published by users in a scenario with a preset theme are collected.

[0033] For example, collect articles posted by users in a shopping circle on a certain topic.

[0034] In step 102, the content of the article and the theme of the scene are scored using a first deep learning model to obtain a first scoring result.

[0035] In some embodiments, scoring the content of the article and the theme of the scene by using the first deep learning model includes: performing word segmentation on the text of the article to obtain corresponding word vectors. Analyzing the theme of the article to obtain the theme vector of the article. Using the first deep learning model to score the word vector, the theme vector and the theme of the scene to obtain a first scoring result.

[0036] In some embodiments, the topic of the article is analyzed by using a Latent Dirichlet allocation (LDA) model to obtain the probability value of the article on each topic, and the topic vector of the article is generated according to the probability value of the article on each topic.

[0037] For example, using the LDA model, the topic of an article can be represented as: 0.1×topic 1+0.6×topic 2+0.15×topic 3+…+0.07×topic 20, thus obtaining a 20-dimensional vector, namely [0.1, 0.6, 0.15,…, 0.07]. This vector is used as the topic vector of the article.

[0038] In some embodiments, the first deep learning model includes multiple sub-models, each of which is of a different type. The word vector, the topic vector, and the topic of the scene are input into the multiple sub-models for scoring, so that each sub-model outputs a corresponding scoring result. The scoring results output by each sub-model are fused to obtain a first scoring result.

[0039] In some embodiments, the multiple sub-models include at least one of an Inception model, an RCNN model, or a Fast Text model.

[0040] It should be noted that the Inception model can obtain text features from different fields of view by using parallel multi-scale convolution kernels, and the use of superimposed convolution layers is conducive to mining deeper text features. The RCNN model can effectively obtain the contextual features of the article. The Fast Text model is simpler and can reduce overfitting.

[0041] Figure 2 The present invention is a flowchart of scoring the content of an article and the theme of a scene according to an embodiment of the present invention.

[0042] like Figure 2As shown in the figure, the text of the article is segmented to obtain the corresponding word vector. The topic of the article is analyzed using the LDA model to obtain the topic vector of the article. The word vector, topic vector and topic of the scene are input into the Inception model, RCNN model and Fast Text model, and the scoring results output by the Inception model, RCNN model and Fast Text model are fused to obtain the first scoring result.

[0043] return Figure 1 In step 103, the statistical features of the text of the article are extracted, and the associated features of the author of the article are extracted.

[0044] In some embodiments, the text statistical features of the article include: at least one of the number of characters, words, paragraphs, or pictures in the article.

[0045] In some embodiments, the author's associated features include the author's attribute information. The author's attribute information includes at least one of the author's level (e.g., ordinary user, celebrity, fashionista, etc.), number of fans, number of followers, the difference between the number of fans and the number of followers (net number of fans), or the ratio of the number of fans to the number of followers (fans-following ratio).

[0046] In some embodiments, the author's associated features also include the author's published article evaluation information. The author's published article evaluation information includes at least one of the total number of articles published by the author, the total number of articles with high-quality tags added to the published articles (the total number of featured articles), the proportion of the total number of articles with high-quality tags added to the published articles (the proportion of featured articles), the average page views PV of the published articles, the number of single visitors UV, the average length of stay, the average number of likes, the average number of comments, or the average number of forwardings.

[0047] In some embodiments, the evaluation information of the author's published article also includes an impact evaluation index. The impact evaluation index is associated with the number of likes, the number of comments, and the number of single visitors to evaluate the degree of dissemination impact of the article.

[0048] For example, the impact evaluation index is (number of likes + 2×number of comments) / (3×number of single visitors).

[0049] In some embodiments, the author's associated features also include the author's trend features. The author's writing level is not static and may rise or fall over time. Therefore, the author's writing status change is evaluated through trend features. For example, the trend feature includes the change in at least one of the author's attribute information or the published article evaluation information within the last T1 day relative to the last T2 day, where T1 <T2。

[0050] For example, the articles can be divided according to their publication time to obtain long-term (all published articles), medium-term (articles published within 90 days) and short-term (articles published within 30 days). By calculating the change in each correlation feature of the short-term relative to the medium-term, or calculating the change in each correlation feature of the medium-term relative to the long-term, the author's change trend characteristics can be determined.

[0051] In step 104, the second deep learning model is used to score the text statistical features of the article and the associated features of the author to obtain a second scoring result.

[0052] In some embodiments, the second deep learning model includes an xgboost model or a lightgbm model.

[0053] In step 105, the first scoring result and the second scoring result are fused to obtain a scoring result of the article.

[0054] In some embodiments, corresponding weights may be set for the first scoring result and the second scoring result as needed. For example, scoring result = 0.4×first scoring result+0.6×second scoring result.

[0055] In some embodiments, manually reviewed articles can be used as training data to add quality labels to the featured articles, thereby training the first deep learning model and the second deep learning model.

[0056] According to the article content scoring method provided by the above embodiment of the present disclosure, during the scoring process, by combining the author's related information and the degree of theme matching between the article topic and the publishing scenario, the scoring can more accurately reflect the quality of the article.

[0057] Figure 3 FIG. 1 is a schematic diagram of the structure of an article content scoring device according to an embodiment of the present disclosure. Figure 3 As shown, the article content scoring device includes a collection module 31 , a first scoring processing module 32 , a second scoring processing module 33 and a scoring module 34 .

[0058] The collection module 31 is configured to collect articles published by users in a scene with a preset theme.

[0059] The first scoring processing module 32 is configured to use the first deep learning model to score the content of the article and the theme of the scene to obtain a first scoring result.

[0060] In some embodiments, the first scoring processing module 32 performs word segmentation processing on the text of the article to obtain corresponding word vectors. The topic of the article is analyzed to obtain the topic vector of the article. The word vector, the topic vector and the topic of the scene are scored using the first deep learning model to obtain a first scoring result.

[0061] In some embodiments, the topic of the article is analyzed by using the LDA model to obtain the probability value of the article on each topic, and the topic vector of the article is generated according to the probability value of the article on each topic.

[0062] In some embodiments, the first deep learning model includes multiple sub-models, each of which is of a different type. The word vector, the topic vector, and the topic of the scene are input into the multiple sub-models for scoring, so that each sub-model outputs a corresponding scoring result. The scoring results output by each sub-model are fused to obtain a first scoring result.

[0063] In some embodiments, the multiple sub-models include at least one of an Inception model, an RCNN model, or a Fast Text model.

[0064] The second scoring processing module 33 is configured to extract text statistical features of the article, extract associated features of the author of the article, and use the second deep learning model to score the text statistical features of the article and the associated features of the author to obtain a second scoring result.

[0065] In some embodiments, the text statistical features of the article include: at least one of the number of characters, words, paragraphs, or pictures in the article.

[0066] In some embodiments, the author's associated features include the author's attribute information. The author's attribute information includes at least one of the author's level (e.g., ordinary user, celebrity, fashionista, etc.), number of fans, number of followers, the difference between the number of fans and the number of followers (net number of fans), or the ratio of the number of fans to the number of followers (fans-following ratio).

[0067] In some embodiments, the author's associated features also include the author's published article evaluation information. The author's published article evaluation information includes at least one of the total number of articles published by the author, the total number of articles with high-quality tags added to the published articles (the total number of featured articles), the proportion of the total number of articles with high-quality tags added to the published articles (the proportion of featured articles), the average page views PV of the published articles, the number of single visitors UV, the average length of stay, the average number of likes, the average number of comments, or the average number of forwardings.

[0068] In some embodiments, the evaluation information of the author's published articles also includes an impact evaluation index. The impact evaluation index is associated with the number of likes, the number of comments, and the number of single visitors to evaluate the degree of dissemination impact of the article. For example, the impact evaluation index is (number of likes + 2×number of comments) / (3×number of single visitors).

[0069] In some embodiments, the author's associated features also include the author's trend features. The author's writing level is not static and may rise or fall over time. Therefore, the author's writing status change is evaluated through trend features. For example, the trend feature includes the change in at least one of the author's attribute information or the published article evaluation information within the last T1 day relative to the last T2 day, where T1 <T2。

[0070] For example, the articles can be divided according to their publication time to obtain long-term (all published articles), medium-term (articles published within 90 days) and short-term (articles published within 30 days). By calculating the change in each correlation feature of the short-term relative to the medium-term, or calculating the change in each correlation feature of the medium-term relative to the long-term, the author's change trend characteristics can be determined.

[0071] In some embodiments, the second deep learning model includes an xgboost model or a lightgbm model.

[0072] The scoring module 34 is configured to fuse the first scoring result and the second scoring result to obtain a scoring result for the article.

[0073] In some embodiments, corresponding weights may be set for the first scoring result and the second scoring result as needed. For example, scoring result = 0.4×first scoring result+0.6×second scoring result.

[0074] Figure 4 FIG. 2 is a schematic diagram of the structure of an article content scoring device according to another embodiment of the present disclosure. Figure 4 As shown, the device includes a memory 41 and a processor 42.

[0075] The memory 41 is used to store instructions. The processor 42 is coupled to the memory 41. The processor 42 is configured to execute the instructions stored in the memory to implement the following steps: Figure 1 The method of any one of the embodiments.

[0076] like Figure 4 As shown, the device also includes a communication interface 43 for information exchange with other devices. At the same time, the device also includes a bus 44, through which the processor 42, the communication interface 43, and the memory 41 communicate with each other.

[0077] The memory 41 may include a high-speed RAM memory, and may also include a non-volatile memory, such as at least one disk memory. The memory 41 may also be a memory array. The memory 41 may also be divided into blocks, and the blocks may be combined into virtual volumes according to certain rules.

[0078] In addition, the processor 42 may be a central processing unit (CPU), or may be an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present disclosure.

[0079] The present disclosure also relates to a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the following are implemented: Figure 1 The method of any one of the embodiments.

[0080] In some embodiments, the functional unit module described above can be implemented as a general-purpose processor, a programmable logic controller (PLC), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components or any appropriate combination thereof for performing the functions described in the present disclosure.

[0081] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0082] The description of the present disclosure is given for the purpose of illustration and description, and is not intended to be exhaustive or to limit the present disclosure to the disclosed form. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present disclosure, and to enable those of ordinary skill in the art to understand the present disclosure and thereby design various embodiments with various modifications suitable for specific uses.

Claims

1. A method for scoring article content, comprising: Collect articles posted by users in scenarios with preset topics; Using a first deep learning model to score the content of the article and the theme of the scene to obtain a first scoring result; Extracting text statistical features of the article; Extracting relevant features of the author of the article; Using a second deep learning model to score the text statistical features of the article and the associated features of the author to obtain a second scoring result; The first scoring result and the second scoring result are merged to obtain the scoring result of the article, and the top article selected according to the scoring result is pinned to the top; The step of using the first deep learning model to score the content of the article and the theme of the scene includes: Perform word segmentation on the text of the article to obtain corresponding word vectors; Analyzing the topic of the article to obtain a topic vector of the article; The word vector, the topic vector, and the topic of the scene are scored using a first deep learning model to obtain the first scoring result, wherein the first scoring result is associated with the degree of match between the topic of the article and the topic of the scene.

2. The method according to claim 1, wherein: The analysis of the topics of the articles included: Analyzing the topics of the article using a latent Dirichlet distribution model to obtain probability values ​​of the article on each topic; A topic vector of the article is generated according to the probability value of the article on each topic.

3. The method according to claim 1, wherein: The first deep learning model includes a plurality of sub-models, and the types of the sub-models are different from each other; Using the first deep learning model to score the word vector, the topic vector, and the topic of the scene includes: Inputting the word vector, the topic vector and the topic of the scene into multiple sub-models for scoring processing, so that each sub-model outputs a corresponding scoring result; The scoring results output by each sub-model are integrated to obtain the first scoring result.

4. The method according to claim 3, wherein: The multiple sub-models include at least one of an Inception model, an RCNN model, or a Fast Text model.

5. The method according to claim 1, wherein: The text statistical features of the article include: at least one of the number of characters, the number of words, the number of paragraphs or the number of pictures in the article.

6. The method according to claim 1, wherein: The associated features of the author include attribute information of the author, wherein the attribute information of the author includes at least one of the author's level, number of fans, number of followers, difference between the number of fans and the number of followers, or ratio of the number of fans to the number of followers.

7. The method according to claim 6, wherein: The author's associated features also include evaluation information of the author's published articles, wherein the author's published article evaluation information includes at least one of the total number of articles published by the author, the total number of articles with high-quality tags added in the published articles, the proportion of the total number of articles with high-quality tags added in the published articles, the average number of views of the published articles, the number of single visitors, the average stay time, the average number of likes, the average number of comments or the average number of reposts.

8. The method according to claim 7, wherein: The evaluation information of the author's published articles also includes an impact evaluation index, wherein the impact evaluation index is associated with the number of likes, the number of comments, and the number of single visitors.

9. The method according to claim 7, wherein: The author's association feature also includes the author's trend feature, which includes the change in at least one of the author's attribute information or published article evaluation information within the past T1 days relative to the past T2 days, where T1 <T2。 10. The method of claim 1, wherein: The second deep learning model includes an xgboost model or a lightgbm model.

11. An article content scoring device, comprising: A collection module, configured to collect articles published by users in a scene with a preset theme; A first scoring processing module is configured to score the content of the article and the theme of the scene using a first deep learning model to obtain a first scoring result, wherein the text of the article is segmented to obtain a corresponding word vector, the theme of the article is analyzed to obtain a theme vector of the article, and the word vector, the theme vector and the theme of the scene are scored using the first deep learning model to obtain the first scoring result, wherein the first scoring result is associated with a degree of matching between the theme of the article and the theme of the scene; A second scoring processing module is configured to extract text statistical features of the article, extract associated features of the author of the article, and score the text statistical features of the article and the associated features of the author using a second deep learning model to obtain a second scoring result; The scoring module is configured to fuse the first scoring result and the second scoring result to obtain the scoring result of the article, and to pin the top article selected according to the scoring result.

12. An article content scoring device, comprising: a memory configured to store instructions; A processor is coupled to the memory, and the processor is configured to execute the method according to any one of claims 1 to 10 based on instructions stored in the memory.

13. A computer-readable storage medium, wherein: The computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the method according to any one of claims 1 to 10 is implemented.

Images