Comment information sorting method, device, storage medium and server

Abstract

The present application discloses a comment information sorting method, device, storage medium and server, which obtains multiple comment information of published content and user interaction information of each comment information, wherein the comment information includes comment content and comment time; determines the popularity value of each comment information according to the comment information, the user interaction information and the published content; determines the time decay value of each comment information according to the comment time; and sorts the multiple comment information according to the popularity value and the time decay value, so as to comprehensively consider factors such as the novelty, popularity and comment time of the comment information, ensure the diversity of the comment sorting list and achieve good sorting effect.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to a comment information sorting method, device, storage medium and server. Background Art

[0002] With the rapid development of Internet technology, in addition to traditional radio and television, the Internet has become a more important channel for obtaining information and disseminating information. People are accustomed to obtaining information from the Internet, and are usually accustomed to posting relevant comments on the Internet to share their experiences or feelings. At the same time, user comments themselves have become an important information. People can obtain more information that is closer to their needs from the comments posted by other users.

[0003] At present, the comment information displayed on the online platform is displayed in descending order according to the generation time of each comment information. Therefore, the valuable comments generated earlier will be displayed at the end. In the process of users browsing the comment information one by one, they may have to read a large number of worthless comment information before obtaining valuable comment information. Obviously, this way of sorting comment information reduces the efficiency of users in obtaining valuable comment information, and the sorting effect is poor. Summary of the Invention

[0004] The embodiments of the present application provide a comment information sorting method, device, storage medium and server, which can prevent early valuable comments from sinking to the bottom and improve the sorting effect of comment information.

[0005] This embodiment of the present application provides a method for sorting review information, including:

[0006] Obtaining multiple comments on published content and user interaction information for each comment, wherein the comment information includes the comment content and comment time;

[0007] Determine the popularity value of each comment information according to the comment information, the user interaction information and the published content;

[0008] Determine a time decay value of each comment information according to the comment time;

[0009] The plurality of comment information are sorted according to the heat value and the time decay value.

[0010] The present application also provides a review information sorting device, including:

[0011] An acquisition module, configured to acquire multiple comments on published content and user interaction information for each comment, wherein the comment information includes the comment content and comment time;

[0012] A first determining module is configured to determine the popularity value of each comment information based on the comment information, the user interaction information, and the published content;

[0013] A second determining module is configured to determine a time decay value of each comment information according to the comment time;

[0014] A sorting module is used to sort the multiple comment information according to the heat value and the time decay value.

[0015] The first determining module specifically includes:

[0016] a first determining unit, configured to determine a content feature of each piece of comment information based on the comment content and the published content;

[0017] a second determining unit, configured to determine an interactive feature of each piece of comment information according to the user interaction information;

[0018] The third determining unit is configured to determine the popularity value of each piece of the comment information according to the interactive feature and the content feature.

[0019] The first determining unit is specifically configured to:

[0020] Determining the relevance between each of the review contents and the published content;

[0021] Determining the number of entity words in each of the review contents and the character length of each of the review contents;

[0022] The relevance, the number of entity words and the character length are used as content features of the corresponding comment information.

[0023] The first determining unit is specifically configured to:

[0024] Determine a first topic vector corresponding to each comment content and a second topic vector corresponding to the published content based on a preset topic model;

[0025] A distribution distance between the second topic vector and each of the first topic vectors is determined to obtain a relevance between the corresponding comment content and the published content.

[0026] The user interaction information includes the number of likes for the comment information, and the number of replies and likes elicited by the comment information. The second determining unit is specifically configured to:

[0027] Determining the complexity of the comment tree of the comment information according to the number of replies elicited by the comment information;

[0028] Determining the number of likes in the comment tree of the comment information according to the number of likes induced by the comment information;

[0029] The number of likes for the comment information, the complexity of the comment tree, and the number of likes for the comment tree are used as interactive features of the corresponding comment information.

[0030] The comment information also includes the commenting user, and the first determining module further includes a fourth determining unit configured to:

[0031] Before the third determining unit determines the popularity value of each comment information according to the interaction feature and the content feature, determining historical interaction information of each network user in a network user set relative to other network users, the network user set including the comment user;

[0032] Determining user characteristics of each commenting user based on the historical interaction information;

[0033] The third determining unit is specifically configured to determine the popularity value of each piece of comment information according to the user characteristics, the interaction characteristics, and the content characteristics.

[0034] The third determining unit is specifically configured to:

[0035] performing logarithmic processing on the user features, the interaction features, and the content features respectively;

[0036] Normalizing the user features, the interaction features, and the content features after logarithmic processing using a minimax method to obtain normalized values;

[0037] According to the preset weighted value, the normalized value corresponding to each comment information is weighted and summed to obtain the corresponding popularity value.

[0038] The historical interaction information includes the total number of historical likes, and the fourth determining unit is specifically configured to:

[0039] Accumulate the total number of likes in history corresponding to each network user to obtain the total number of likes on the network;

[0040] Sorting the network users according to the total number of likes in history corresponding to each network user;

[0041] Determining the user level of each commenting user based on the total number of likes on the network and the ranked network users;

[0042] According to the user level of each commenting user, a corresponding like weight value is determined, and the like weight value is used as a user feature.

[0043] Wherein, the historical interaction information includes historical likes or replies, and the fourth determining unit is specifically configured to:

[0044] Constructing a network node graph using the historical likes or reply relationships as edges and the network users as nodes;

[0045] According to a preset user rating algorithm and the network node graph, a node weight value of each commenting user is determined, and the node weight value is used as a user feature.

[0046] The sorting module is specifically used for:

[0047] Calculate the product of the popularity value and the time decay value corresponding to each comment information to obtain the recommendation degree;

[0048] The plurality of review information are sorted according to the numerical values of the recommendation degrees.

[0049] The comment information sorting device further includes an adjustment display module for:

[0050] After the sorting module sorts the plurality of comment information according to the popularity value and the time decay value, a low-quality classification label of each comment information is determined using a preset classification model;

[0051] Adjusting the positions of the sorted plurality of review information according to the low-quality classification labels;

[0052] The adjusted plurality of comment information is displayed on the comment interface of the published article.

[0053] An embodiment of the present application also provides a computer-readable storage medium, in which a plurality of instructions are stored. The instructions are suitable for being loaded by a processor to execute any of the above-mentioned comment information sorting methods.

[0054] An embodiment of the present application also provides a server, including a processor and a memory, wherein the processor is electrically connected to the memory, the memory is used to store instructions and data, and the processor is used to execute the steps in any of the above-mentioned comment information sorting methods.

[0055] The comment information sorting method, device, storage medium and server provided by the present application obtain multiple comment information of published content and user interaction information of each comment information, the comment information including comment content and comment time, and determine the heat value of each comment information based on the comment information, the user interaction information and the published content, and then determine the time decay value of each comment information based on the comment time, and sort the multiple comment information according to the heat value and time decay value, so as to comprehensively consider factors such as the novelty, heat and comment time of the comment information, ensure the diversity of the comment sorting list, not only avoid early valuable comments from sinking to the bottom, but also avoid new comments from having sufficient exposure opportunities, thereby improving the comment sorting effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The following detailed description of the specific embodiments of the present application in conjunction with the accompanying drawings will make the technical solutions and other beneficial effects of the present application apparent.

[0057] Figure 1 A schematic diagram of a scenario of a review information ranking system provided in an embodiment of the present application.

[0058] Figure 2 A flowchart of the comment information sorting method provided in an embodiment of the present application.

[0059] Figure 3 Another flowchart of the comment information sorting method provided in an embodiment of the present application.

[0060] Figure 4 A schematic diagram showing the review information provided in an embodiment of the present application.

[0061] Figure 5 A schematic diagram of the user likes distribution and the number of users provided in an embodiment of the present application.

[0062] Figure 6 A schematic diagram of a network node diagram provided in an embodiment of the present application.

[0063] Figure 7 A schematic diagram of the feature classification of comment information provided in an embodiment of the present application.

[0064] Figure 8 A schematic diagram of the components of the computing module provided in an embodiment of the present application.

[0065] Figure 9 A schematic diagram of the structure of the comment information sorting device provided in an embodiment of the present application.

[0066] Figure 10 Another structural diagram of the comment information sorting device provided in an embodiment of the present application.

[0067] Figure 11 A schematic diagram of the structure of the server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0069] Embodiments of the present application provide a comment information sorting method, device, storage medium, and server.

[0070] See also Figure 1 , Figure 1 This is a scenario diagram of a comment information sorting system. The comment information sorting system may include any one of the comment information sorting devices provided in the embodiments of the present application. The comment information sorting device may be integrated in a server. The server may be a backend server of a network platform. The network platform may be mainly used to display multimedia products to network users, such as short videos, videos, articles, etc.

[0071] The server can obtain multiple comment information of the published content and user interaction information of each comment information, the comment information including the comment content and the comment time; determine the heat value of each comment information based on the comment information, the user interaction information and the published content; determine the time decay value of each comment information based on the comment time; and sort the multiple comment information according to the heat value and the time decay value.

[0072] Published content primarily refers to content published by online users on online platforms, such as published videos or articles. Online users refer to users who have registered and / or logged in to the online platform. Comment information refers to information generated when any online user comments on published content, primarily in text form. User interaction information refers to information generated when any online user interacts with published content, including likes and replies.

[0073] For example, see Figure 1The server is the background server of the video playback platform. For the published xxx TV series video, network users can comment and interact with it on the display interface of the xxx TV series video. For example, there are 10 comment messages, and a comment message is "User A commented 25 minutes ago: What does it mean? Only 5 episodes have been updated." The user interaction information of the comment message includes 8 likes and 3 replies from other network users. At this time, the server can obtain all the comment information and user interaction information under the xxx TV series video, and determine the heat value of each comment information based on the comment information, user interaction information and xxx TV series video. At the same time, the time decay value is determined according to the comment time. Then, the 10 comment messages are sorted and displayed according to the heat value and time decay value. For example, the comment information of user A, the comment information of user C...the comment information of user J, etc. can be displayed on the display interface in sequence.

[0074] like Figure 2 As shown, Figure 2 This is a flow chart of a comment information sorting method provided in an embodiment of the present application. The comment information sorting method is applied to a server. The specific process may be as follows:

[0075] S101. Acquire multiple comment information of published content and user interaction information of each comment information, wherein the comment information includes the comment content and comment time.

[0076] Published content primarily refers to content published by online users on online platforms, such as published videos or articles. Online users refer to users who have registered and / or logged in to the online platform. Comment information refers to information generated when any online user comments on published content, primarily in text form. User interaction information refers to information generated when any online user interacts with published content, including likes, replies, and / or reposts.

[0077] S102. Determine the popularity value of each comment information based on the comment information, the user interaction information, and the published content.

[0078] The calculation of the heat value comprehensively considers multiple reference features, such as content features and interaction features. The content feature mainly reflects the comment content itself, and the interaction feature mainly reflects the interaction of each network user with respect to the comment. Each reference feature may include at least one numerical value.

[0079] For example, see Figure 3 , the above step S102 may specifically include:

[0080] S1021. Determine the content characteristics of each comment information based on the comment content and the published content.

[0081] For example, the above step S1021 may specifically include:

[0082] Determine the relevance between each comment and the published content;

[0083] Determine the number of entity words in each comment and the character length of each comment;

[0084] The relevance, the number of entity words and the character length are used as content features of the corresponding comment information.

[0085] Among them, the relevance reflects the similarity between the comment content and the published content. Generally, the higher the similarity, the greater the value of the comment content. The number of entity words reflects the fullness of the comment content. The more entity words, the fuller the content and the greater the value. Specifically, the number of entity words in the comment content can be identified by the NER (Named Entity Recognition) model. The NER model can be a combination of multiple models, such as LSTM (Long Short-Term Memory) model + CRF (Conditional Random Field) model. When calculating the character length of the comment content, considering that the contribution value of repeated content is not great, the comment content can be deduplicated first, and repeated phrases or phrases can be removed. Then, the number of characters in the remaining content can be counted to obtain the character length.

[0086] Furthermore, the above step of “determining the relevance between each comment and the published content” may specifically include:

[0087] Determine a first topic vector corresponding to each comment content and a second topic vector corresponding to the published content based on a preset topic model;

[0088] A distribution distance between the second topic vector and each of the first topic vectors is determined to obtain a relevance between the corresponding comment content and the published content.

[0089] Among them, when the published content is non-textual content such as video or animation, the corresponding text content can be determined first, such as using the video introduction, animation introduction, etc. as the text content of the published content, or recognizing the voice, subtitles, etc. of the video or animation, and using the recognized content as the text content.

[0090] The preset topic model may include an LDA (Latent Dirichlet Allocation) model. After obtaining the text content of the published content, the LDA model can be used to identify the text content and comment content of the published content respectively to obtain their respective corresponding topic vectors (i.e., the first topic vector and the second topic vector). Afterwards, the distribution distance between the first topic vector and the second topic vector can be calculated by the JS divergence (Jensen-Shanno divergence, JSD) algorithm to determine the relevance between the published content and the comment content. Alternatively, other similarity algorithms can be used to calculate the relevance between the two, such as the cosine distance.

[0091] Among them, for the first topic vector P and the second topic vector Q, the distribution distance JSD(P||Q) between the two is calculated as follows:

[0092]

[0093] M=(P+Q) / 2,

[0094] Where D(P||Q) is the Kullback-Leible divergence (KLD) or relative entropy, D(P||M) is the PM divergence, and D(Q||M) is the QM divergence.

[0095] S1022. Determine the interactive features of each comment information based on the user interactive information.

[0096] For example, the user interaction information may include the number of likes for the comment information, and the number of replies and likes elicited by the comment information. In this case, the above step S1022 specifically includes:

[0097] Determine the complexity of the comment tree of the comment information according to the number of replies elicited by the comment information;

[0098] Determine the number of likes in the comment tree of the comment information according to the number of likes induced by the comment information;

[0099] The number of likes for the comment information, the complexity of the comment tree, and the number of likes for the comment tree are used as interactive features of the corresponding comment information.

[0100] Among them, the number of replies (or likes) generated by the comment information includes the number of all reply operations (or like operations) under the published content, including the number of replies (likes) to the comment information by network users, and the number of replies (likes) to the reply information by network users.

[0101] For example, see Figure 4Suppose there are 10 comments under a published article, among which other network users replied to a certain comment a 4 times and liked it 2 times, and other network users replied to one of the replies 3 times and liked it 3 times. Then the number of likes for the comment is 2, the number of likes induced by the comment is 2+3=5, and the number of replies induced by the comment is 4+3=7.

[0102] A comment tree refers to the tree structure formed by the interactive operations of multiple network users under a single comment message. For example, a single comment message may have multiple likes and replies, some replies may have likes and replies, and replies to replies may have likes and replies. This hierarchical form forms the comment tree. The complexity of the comment tree reflects the richness of the interactive operations elicited by a single comment message. Generally, the greater the complexity of the comment tree, the more interactive operations such as likes, replies, and reposts that network users participate in. The complexity of the comment tree (or the number of likes in the comment tree) can be directly equal to the number of replies (or likes) elicited by the comment message. Of course, the complexity of the comment tree can also be determined in combination with other information, such as the number of reposts, the popularity of the replying user, etc., and the complexity of the comment tree can be determined through weighted methods.

[0103] It should be noted that, in addition to using some values related to user likes as interactive features, some values related to other interactive operations can also be used as interactive features, such as the number of reposts.

[0104] S1023. Determine the popularity value of each comment information based on the interactive feature and the content feature.

[0105] The popularity value can be calculated directly based on the interactive features and content features, or can be calculated in combination with other dimensional features, such as user features. In this case, the comment information can also include the commenting user. Before the above step S1023, the comment information sorting method can also include the following steps S1024-S1025, wherein:

[0106] S1024. Determine historical interaction information of each network user with respect to other network users in a network user set, where the network user set includes the commenting user.

[0107] The network user set is the collection of all network users registered and / or logged in on the network platform, including users who posted articles, videos, or animations, commenters who participated in comments, and users who participated in interactions (such as likes and replies). Historical interaction information mainly refers to information generated when network users interact with each other, such as likes, replies, and reposts.

[0108] S1025. Determine the user characteristics of each commenting user based on the historical interaction information.

[0109] Among them, the user feature mainly reflects the commenting user itself, which may include at least one numerical value. Different numerical values are obtained from different measurement angles. For example, the user feature can be obtained by considering the interactive correlation between users, or by considering the user's own interactive behavior.

[0110] For example, when the user's own interactive behavior is considered to obtain user characteristics, the historical interactive information may include the total number of historical likes. The above step S1025 may specifically include:

[0111] Accumulate the total number of likes in history for each network user to get the total number of likes on the network;

[0112] Sort the network users according to the total number of likes they have received in history.

[0113] Determine the user level of each commenting user based on the total number of likes on the network and the ranked users of the network;

[0114] According to the user level of each commenting user, the corresponding like weight value is determined, and the like weight value is used as the user feature.

[0115] In this embodiment, taking into account the different habits of liking behaviors of different network users, the likes of users who cherish the behavior of liking are often more valuable than those of users who frequently like. Based on this, the like behaviors of all network users in the historical period can be recorded, and the total number of historical likes of each network user and the total number of network likes can be counted. Afterwards, all network users are sorted according to the rule that the more likes, the lower the ranking of the network user, and the total number of network likes is divided into N equal parts according to the sorting order. Different user levels are set for network users in different equal parts, and different like weight values are set for different user levels. Generally, the network users corresponding to the higher the ranking equal parts are, the higher the user level is, and the greater the like weight value is.

[0116] See Bar Chart Figure 5 , assuming that the total number of likes on the network is divided into 10 equal parts, such as 10%, 20%...100%, that is, divided into 10 user levels. The higher the user level, the fewer likes the network user has. Figure 5 It can be seen that, in order from low to high user levels, the number of network users corresponding to each equal division is: 21, 70, 173...100000. It is easy to see that the higher the user level, the fewer the corresponding network users.

[0117] In addition, when the interaction correlation between users is considered to obtain user features, the historical interaction information may include historical likes or replies. The above step S1025 may specifically include:

[0118] Construct a network node graph using the historical likes or replies as edges and the network users as nodes;

[0119] According to the preset user level algorithm and the network node graph, the node weight value of each comment user is determined, and the node weight value is used as the user feature.

[0120] Among them, a network node graph can be constructed with all network users on the network platform as nodes and the likes or reply relationships between network users as edges. For example, suppose there are network users AF, among which network user A has liked B, C, and D, B has liked C, C has liked D, D has liked B, C, and F, and E has liked F. If a network node graph is constructed with like relationships as edges and network users as nodes, the resulting network node graph is Figure 6 .

[0121] The preset user ranking algorithm may be a PeopleRank algorithm, which can be used to calculate the weight value of each node in the network node graph, that is, to obtain the weight value of each network user, from which the weight value of the commenting user can be selected as the user feature.

[0122] At the same time, the above-mentioned step S1023 may specifically include: determining the popularity value of each comment information according to the user characteristics, the interaction characteristics and the content characteristics.

[0123] Furthermore, the above step of “determining the popularity value of each comment information based on the user characteristics, the interaction characteristics, and the content characteristics” may specifically include:

[0124] Performing logarithmic processing on the user feature, the interaction feature, and the content feature respectively;

[0125] Using the minimax method, normalizing the user feature, the interaction feature, and the content feature after logarithmic processing to obtain normalized values;

[0126] According to the preset weighted value, the normalized value corresponding to each comment information is weighted and summed to obtain the corresponding popularity value.

[0127] In this embodiment, the calculation formula for the popularity value H of any comment information can be:

[0128] H=∑ i w i *min max regressionlog(factor i )

[0129] Among them, factor i is the i-th feature value, which is any one of the feature values of all the above features corresponding to a single comment information, for example, see Figure 7 The features of a single comment can be divided into three types: content features, interaction features and user features. Among them, content features include three feature values: relevance, number of entity words and character length. Interaction features include three feature values: number of likes for comment information, complexity of comment tree and number of likes for comment tree. User features include three feature values: node weight value or like weight value. i For any of these eigenvalues, the calculation method of each eigenvalue can refer to the above steps.

[0130] log(factor i ) is the logarithm of the ith eigenvalue, min max regression log(factor i ) is to perform the minimax method on the logarithm of the ith eigenvalue, which is obtained by obtaining the logarithm of the ith eigenvalue of all comment information, and based on the logarithm of all the ith eigenvalues, the logarithm of the ith eigenvalue of a single comment information is processed to unify the dimension of the eigenvalue of a single dimension. i It is the preset weighted value of the i-th eigenvalue, which can be set manually. Different preset weighted values can be set for different eigenvalues of the same product, and the preset weighted values for the same eigenvalue of different products can be set to different values.

[0131] S103. Determine the time decay value of each comment message according to the comment time.

[0132] Among them, the difference between the current time and the comment time can be calculated first, and the time decay value can be determined based on the difference, for example, the time decay value gravity time The calculation formula can be as follows:

[0133]

[0134] g=e -△t*α ,

[0135] Among them, △t is the difference between the current time and the comment time, and α is a fixed value set artificially.

[0136] S104. Sort the multiple comment information according to the popularity value and time decay value.

[0137] The above step S104 may specifically include:

[0138] Calculate the product of the popularity value and time decay value corresponding to each comment information to obtain the recommendation degree;

[0139] The plurality of review information are sorted according to the numerical values of the recommendation degrees.

[0140] In this embodiment, these comment information can be sorted in descending order of recommendation degree. Since the calculation of the recommendation degree combines multiple feature dimensions, the diversity of the comment sorting list can be ensured, which not only prevents early valuable comments from sinking to the bottom, but also prevents new comments from having sufficient exposure opportunities.

[0141] The calculation formula for the recommendation score of a single comment can be:

[0142] Score=H*gravity time , where H is the above heat value, gravity time is the above time decay value.

[0143] In addition, after the above step S104, the comment information sorting method may further include:

[0144] S105. Determine a low-quality classification label for each comment using a preset classification model;

[0145] S106. Adjust the positions of the sorted plurality of review information according to the low-quality classification labels;

[0146] S107. Display the adjusted plurality of comment information on the comment interface of the published article.

[0147] The preset classification model can be a BERT (Bidirectional Encoder Representations from Transformers) model. Low-quality classification labels can include advertising comments, abusive comments, and vulgar comments. Different low-quality classification labels can be set with different adjustment ranges. The position of the sorted comment information can be optimized and adjusted by the respective adjustment ranges to reduce the ranking position of the comment information with low-quality content. In other words, please refer to Figure 8 The comment information sorting method in this embodiment may include three calculation modules: a feature calculation module, a rough sorting module and a fine sorting optimization module, wherein the feature calculation module is used to calculate the various feature values mentioned above, the rough sorting module is used to calculate the recommendation degree according to the feature value, and roughly sort the comment information based on the recommendation degree, and the fine sorting optimization module is used to determine the low-quality classification label, and adjust the position of the roughly sorted comment information based on the low-quality classification label.

[0148] From the above, it can be seen that the comment information sorting method provided by this application obtains multiple comment information of published content and user interaction information of each comment information, the comment information includes comment content and comment time, and determines the heat value of each comment information based on the comment information, the user interaction information and the published content, and then determines the time decay value of each comment information based on the comment time, and sorts the multiple comment information according to the heat value and time decay value, so as to comprehensively consider factors such as the novelty, heat and comment time of the comment information, ensure the diversity of the comment sorting list, not only avoid early valuable comments from sinking to the bottom, but also avoid new comments from having sufficient exposure opportunities, thereby improving the comment sorting effect.

[0149] According to the method described in the above embodiment, this embodiment will be further described from the perspective of the comment information sorting device. The comment information sorting device can be implemented as an independent entity or integrated in a server. The server can be the background server of the network platform. The network platform can be mainly used to display multimedia products to network users, such as short videos, videos, articles, etc.

[0150] See also Figure 9 , Figure 9 The review information sorting device provided by the embodiment of the present application is specifically described. The review information sorting device may include: an acquisition module 10, a first determination module 20, a second determination module 30, and a sorting module 40, wherein:

[0151] (1) Get module 10

[0152] The acquisition module 10 is used to acquire multiple comment information of the published content and user interaction information of each comment information, wherein the comment information includes the comment content and the comment time.

[0153] Published content primarily refers to content published by online users on online platforms, such as published videos or articles. Online users refer to users who have registered and / or logged in to the online platform. Comment information refers to information generated when any online user comments on published content, primarily in text form. User interaction information refers to information generated when any online user interacts with published content, including likes, replies, and / or reposts.

[0154] (2) First determination module 20

[0155] The first determining module 20 is configured to determine the popularity value of each comment information according to the comment information, the user interaction information and the published content.

[0156] The calculation of the heat value comprehensively considers multiple reference features, such as content features and interaction features. The content feature mainly reflects the comment content itself, and the interaction feature mainly reflects the interaction of each network user with respect to the comment. Each reference feature may include at least one numerical value.

[0157] For example, see Figure 10 The first determining module 20 may specifically include a first determining unit 21, a second determining unit 22, and a third determining unit 23, wherein:

[0158] A first determining unit 21 is configured to determine a content feature of each comment information according to the comment content and the published content;

[0159] For example, the first determining unit 21 is specifically configured to:

[0160] Determine the relevance between each comment and the published content;

[0161] Determine the number of entity words in each comment and the character length of each comment;

[0162] The relevance, the number of entity words and the character length are used as content features of the corresponding comment information.

[0163] Among them, the relevance reflects the similarity between the comment content and the published content. Generally, the higher the similarity, the greater the value of the comment content. The number of entity words reflects the fullness of the comment content. The more entity words, the fuller the content and the greater the value. Specifically, the number of entity words in the comment content can be identified by the NER (Named Entity Recognition) model. The NER model can be a combination of multiple models, such as LSTM (Long Short-Term Memory) model + CRF (Conditional Random Field) model. When calculating the character length of the comment content, considering that the contribution value of repeated content is not great, the comment content can be deduplicated first, and repeated phrases or phrases can be removed. Then, the number of characters in the remaining content can be counted to obtain the character length.

[0164] Furthermore, the first determining unit 21 is specifically configured to:

[0165] Determine a first topic vector corresponding to each comment content and a second topic vector corresponding to the published content based on a preset topic model;

[0166] A distribution distance between the second topic vector and each of the first topic vectors is determined to obtain a relevance between the corresponding comment content and the published content.

[0167] Among them, when the published content is non-textual content such as video or animation, the corresponding text content can be determined first, such as using the video introduction, animation introduction, etc. as the text content of the published content, or recognizing the voice, subtitles, etc. of the video or animation, and using the recognized content as the text content.

[0168] The preset topic model may include an LDA (Latent Dirichlet Allocation) model. After obtaining the text content of the published content, the LDA model can be used to identify the text content and comment content of the published content respectively to obtain their respective corresponding topic vectors (i.e., the first topic vector and the second topic vector). Afterwards, the distribution distance between the first topic vector and the second topic vector can be calculated using the JS divergence (Jensen-Shanno divergence, JSD) algorithm to determine the relevance between the published content and the comment content. Alternatively, other similarity algorithms can be used to calculate the relevance between the two, such as the cosine distance.

[0169] Among them, for the first topic vector P and the second topic vector Q, the distribution distance JSD(P||Q) between the two is calculated as follows:

[0170]

[0171] M=(P+Q) / 2,

[0172] Where D(P||Q) is the Kullback-Leible divergence (KLD) or relative entropy, D(P||M) is the PM divergence, and D(Q||M) is the QM divergence.

[0173] The second determining unit 22 is configured to determine an interactive feature of each piece of comment information according to the user interaction information.

[0174] For example, the user interaction information includes the number of likes for the comment information, and the number of replies and likes elicited by the comment information. The second determining unit 22 is specifically configured to:

[0175] Determine the complexity of the comment tree of the comment information according to the number of replies elicited by the comment information;

[0176] Determine the number of likes in the comment tree of the comment information according to the number of likes induced by the comment information;

[0177] The number of likes for the comment information, the complexity of the comment tree, and the number of likes for the comment tree are used as interactive features of the corresponding comment information.

[0178] Among them, the number of replies (or likes) generated by the comment information includes the number of all reply operations (or like operations) under the published content, including the number of replies (likes) to the comment information by network users, and the number of replies (likes) to the reply information by network users.

[0179] For example, see Figure 4 Suppose there are 10 comments under a published article, among which other network users replied to a certain comment a 4 times and liked it 2 times, and other network users replied to one of the replies 3 times and liked it 3 times. Then the number of likes for the comment is 2, the number of likes induced by the comment is 2+3=5, and the number of replies induced by the comment is 4+3=7.

[0180] A comment tree refers to the tree structure formed by the interactive operations of multiple network users under a single comment message. For example, a single comment message may have multiple likes and replies, some replies may have likes and replies, and replies to replies may have likes and replies. This hierarchical form forms the comment tree. The complexity of the comment tree reflects the richness of the interactive operations elicited by a single comment message. Generally, the greater the complexity of the comment tree, the more interactive operations such as likes, replies, and reposts that network users participate in. The complexity of the comment tree (or the number of likes in the comment tree) can be directly equal to the number of replies (or likes) elicited by the comment message. Of course, the complexity of the comment tree can also be determined in combination with other information, such as the number of reposts, the popularity of the replying user, etc., and the complexity of the comment tree can be determined through weighted methods.

[0181] It should be noted that, in addition to using some values related to user likes as interactive features, some values related to other interactive operations can also be used as interactive features, such as the number of reposts.

[0182] The third determining unit 23 is configured to determine the popularity value of each comment information according to the interaction feature and the content feature.

[0183] The popularity value may be calculated directly based on the interaction features and content features, or may be calculated in combination with other dimensional features, such as user features. In this case, the comment information may also include the commenting user. The first determining module 20 further includes a fourth determining unit 24 for:

[0184] Before the third determining unit 23 determines the popularity value of each comment information according to the interaction feature and the content feature, the following steps S1024-S1025 are performed, wherein:

[0185] S1024. Determine historical interaction information of each network user with respect to other network users in a network user set, where the network user set includes the commenting user.

[0186] The network user set is the collection of all network users registered and / or logged in on the network platform, including users who posted articles, videos, or animations, commenters who participated in comments, and users who participated in interactions (such as likes and replies). Historical interaction information mainly refers to information generated when network users interact with each other, such as likes, replies, and reposts.

[0187] S1025. Determine the user characteristics of each commenting user based on the historical interaction information.

[0188] Among them, the user feature mainly reflects the commenting user itself, which may include at least one numerical value. Different numerical values are obtained from different measurement angles. For example, the user feature can be obtained by considering the interactive correlation between users, or by considering the user's own interactive behavior.

[0189] For example, when the user's own interactive behavior is considered to obtain user characteristics, the historical interaction information may include the total number of historical likes. When executing the above step S1025, the fourth determining unit 24 may be specifically used to:

[0190] Accumulate the total number of likes in history for each network user to get the total number of likes on the network;

[0191] Sort the network users according to the total number of likes they have received in history.

[0192] Determine the user level of each commenting user based on the total number of likes on the network and the ranked users of the network;

[0193] According to the user level of each commenting user, the corresponding like weight value is determined, and the like weight value is used as the user feature.

[0194] In this embodiment, taking into account the different habits of liking behaviors of different network users, the likes of users who cherish the behavior of liking are often more valuable than those of users who frequently like. Based on this, the like behaviors of all network users in the historical period can be recorded, and the total number of historical likes of each network user and the total number of network likes can be counted. Afterwards, all network users are sorted according to the rule that the more likes, the lower the ranking of the network user, and the total number of network likes is divided into N equal parts according to the sorting order. Different user levels are set for network users in different equal parts, and different like weight values are set for different user levels. Generally, the network users corresponding to the higher the ranking equal parts are, the higher the user level is, and the greater the like weight value is.

[0195] See Bar Chart Figure 5, assuming that the total number of likes on the network is divided into 10 equal parts, such as 10%, 20%...100%, that is, divided into 10 user levels. The higher the user level, the fewer likes the network user has. Figure 5 It can be seen that, in order from low to high user levels, the number of network users corresponding to each equal division is: 21, 70, 173...100000. It is easy to see that the higher the user level, the fewer the corresponding network users.

[0196] In addition, when considering the interaction correlation between users to obtain user features, the historical interaction information may include historical likes or replies. When executing step S1025, the fourth determining unit 24 may be specifically configured to:

[0197] Construct a network node graph using the historical likes or replies as edges and the network users as nodes;

[0198] According to the preset user level algorithm and the network node graph, the node weight value of each comment user is determined, and the node weight value is used as the user feature.

[0199] Among them, a network node graph can be constructed with all network users on the network platform as nodes and the likes or reply relationships between network users as edges. For example, suppose there are network users AF, among which network user A has liked B, C, and D, B has liked C, C has liked D, D has liked B, C, and F, and E has liked F. If a network node graph is constructed with like relationships as edges and network users as nodes, the resulting network node graph is Figure 6 .

[0200] The preset user ranking algorithm may be a PeopleRank algorithm, which can be used to calculate the weight value of each node in the network node graph, that is, to obtain the weight value of each network user, from which the weight value of the commenting user can be selected as the user feature.

[0201] At the same time, when executing the above step S1023, the third determining unit 23 can be specifically used to determine the popularity value of each comment information according to the user feature, the interaction feature and the content feature.

[0202] Furthermore, the third determining unit 23 is specifically configured to:

[0203] Performing logarithmic processing on the user feature, the interaction feature, and the content feature respectively;

[0204] Using the minimax method, normalizing the user feature, the interaction feature, and the content feature after logarithmic processing to obtain normalized values;

[0205] According to the preset weighted value, the normalized value corresponding to each comment information is weighted and summed to obtain the corresponding popularity value.

[0206] In this embodiment, the calculation formula for the popularity value H of any comment information can be:

[0207] H=∑ i w i *min max regression log(factor i )

[0208] Among them, factor i is the i-th feature value, which is any one of the feature values of all the above features corresponding to a single comment information, for example, see Figure 7 The features of a single comment can be divided into three types: content features, interaction features and user features. Content features include relevance, number of entity words and character length. Interaction features include the number of likes for the comment, the complexity of the comment tree and the number of likes for the comment tree. User features include node weight or like weight. i For any of these eigenvalues, the calculation method of each eigenvalue can refer to the above steps.

[0209] log(factor i ) is the logarithm of the ith eigenvalue, min max regression log(factor i ) is to perform the minimax method on the logarithm of the ith eigenvalue, which is obtained by obtaining the logarithm of the ith eigenvalue of all comment information, and based on the logarithm of all the ith eigenvalues, the logarithm of the ith eigenvalue of a single comment information is processed to unify the dimension of the eigenvalue of a single dimension. i It is the preset weighted value of the i-th eigenvalue, which can be set manually. Different preset weighted values can be set for different eigenvalues of the same product, and the preset weighted values for the same eigenvalue of different products can be set to different values.

[0210] (3) Second determination module 30

[0211] The second determining module 30 is configured to determine a time decay value of each comment message according to the comment time.

[0212] Among them, the difference between the current time and the comment time can be calculated first, and the time decay value can be determined based on the difference, for example, the time decay value gravity time The calculation formula can be as follows:

[0213]

[0214] g=e -△t*α ,

[0215] Among them, △t is the difference between the current time and the comment time, and α is a fixed value set artificially.

[0216] (4) Sorting module 40

[0217] The sorting module 40 is used to sort the plurality of comment information according to the popularity value and the time decay value.

[0218] The sorting module 40 is specifically configured to:

[0219] Calculate the product of the popularity value and time decay value corresponding to each comment information to obtain the recommendation degree;

[0220] The plurality of review information are sorted according to the numerical values of the recommendation degrees.

[0221] In this embodiment, the review information can be sorted in descending order of recommendation degree. Since the calculation of the recommendation degree combines multiple feature dimensions, the diversity of the review ranking list can be ensured, which not only prevents early valuable reviews from sinking to the bottom, but also prevents new reviews from having sufficient exposure opportunities.

[0222] The calculation formula for the recommendation score of a single comment can be:

[0223] Score=H*gravity time , where H is the above heat value, gravity time is the above time decay value.

[0224] In addition, the comment information sorting device further includes an adjustment display module 50 for:

[0225] After the sorting module 40 sorts the plurality of review information according to the popularity value and the time decay value, a low-quality classification label of each review information is determined using a preset classification model;

[0226] Adjusting the positions of the sorted plurality of review information according to the low-quality classification labels;

[0227] The adjusted plurality of comment information is displayed on the comment interface of the published article.

[0228] The preset classification model can be a BERT (Bidirectional Encoder Representations from Transformers) model. Low-quality classification labels can include advertising comments, abusive comments, and vulgar comments. Different low-quality classification labels can be set with different adjustment ranges. The position of the sorted comment information can be optimized and adjusted by the respective adjustment ranges to reduce the ranking position of the comment information with low-quality content. In other words, please refer to Figure 8 The comment information sorting method in this embodiment may include three calculation modules: a feature calculation module, a rough sorting module and a fine sorting optimization module, wherein the feature calculation module is used to calculate the various feature values mentioned above, the rough sorting module is used to calculate the recommendation degree according to the feature value, and roughly sort the comment information based on the recommendation degree, and the fine sorting optimization module is used to determine the low-quality classification label, and adjust the position of the roughly sorted comment information based on the low-quality classification label.

[0229] In specific implementation, the above units can be implemented as independent entities, or can be arbitrarily combined to be implemented as the same or several entities. The specific implementation of the above units can be found in the previous method embodiments and will not be repeated here.

[0230] As can be seen from the above, the comment information sorting device provided by this embodiment obtains multiple comment information of the published content and user interaction information of each comment information through the acquisition module 10. The comment information includes the comment content and the comment time. The first determination module 20 determines the popularity value of each comment information based on the comment information, the user interaction information and the published content. Then the second determination module 30 determines the time decay value of each comment information based on the comment time. The sorting module 40 sorts the multiple comment information according to the popularity value and the time decay value, so as to comprehensively consider factors such as the novelty, popularity and comment time of the comment information, ensure the diversity of the comment sorting list, not only avoid early valuable comments from sinking to the bottom, but also avoid new comments from having sufficient exposure opportunities, thereby improving the comment sorting effect.

[0231] Correspondingly, an embodiment of the present invention further provides a comment information sorting system, including any one of the comment information sorting devices provided in the embodiments of the present invention. The comment information sorting device can be integrated into a server.

[0232] The server may obtain multiple comment information of the published content and user interaction information of each comment information, wherein the comment information includes the comment content and comment time;

[0233] Determine the popularity value of each comment information based on the comment information, the user interaction information and the published content;

[0234] Determine the time decay value of each comment information according to the comment time;

[0235] The plurality of comment pieces of information are sorted according to the heat value and the time decay value.

[0236] The specific implementation of each of the above devices can be found in the previous embodiments and will not be described again here.

[0237] Since the comment information sorting system can include any comment information sorting device provided in the embodiments of the present invention, it can achieve the beneficial effects that can be achieved by any comment information sorting device provided in the embodiments of the present invention. Please refer to the previous embodiments for details and will not be repeated here.

[0238] Accordingly, the embodiment of the present application also provides a server, such as Figure 11 As shown, the server may include components such as a radio frequency (RF) circuit 601, a memory 602 including one or more computer-readable storage media, an input unit 603, a display unit 604, a sensor 605, an audio circuit 606, a wireless fidelity (WiFi) module 607, a processor 608 including one or more processing cores, and a power supply 609. It will be understood by those skilled in the art that Figure 11 The server structure shown in the figure does not constitute a limitation on the server, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0239] The RF circuit 601 can be used to receive and send signals during information transmission or calls. Specifically, after receiving downlink information from the base station, it is handed over to one or more processors 608 for processing; in addition, uplink data is sent to the base station. Generally, the RF circuit 601 includes but is not limited to an antenna, at least one amplifier, a tuner, one or more oscillators, a subscriber identity module (SIM) card, a transceiver, a coupler, a low noise amplifier (LNA), a duplexer, etc. In addition, the RF circuit 601 can also communicate with the network and other devices via wireless communication. This wireless communication can use any communication standard or protocol, including but not limited to Global System of Mobile Communications (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, Short Messaging Service (SMS), etc.

[0240] The memory 602 can be used to store software programs and modules. The processor 608 executes various functional applications and comment information sorting by running the software programs and modules stored in the memory 602. The memory 602 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the server (such as audio data, a phone book, etc.), etc. In addition, the memory 602 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. Accordingly, the memory 602 may also include a memory controller to provide the processor 608 and the input unit 603 with access to the memory 602.

[0241] The input unit 603 can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical, or trackball signal input related to user settings and function control. Specifically, in one embodiment, the input unit 603 may include a touch-sensitive surface and other input devices. A touch-sensitive surface, also known as a touch display or touchpad, can detect user touch operations on or near it (for example, operations performed by a user using a finger, stylus, or any other suitable object or accessory on or near the touch-sensitive surface) and drive corresponding connected devices according to a pre-set program. Optionally, the touch-sensitive surface may include a touch detection device and a touch controller. The touch detection device detects the user's touch direction and detects signals generated by the touch operation, transmitting the signals to the touch controller. The touch controller receives the touch information from the touch detection device, converts it into touch point coordinates, and then sends it to the processor 608. It can also receive and execute commands from the processor 608. In addition, touch-sensitive surfaces can be implemented using various types, such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch-sensitive surface, the input unit 603 may also include other input devices. Specifically, other input devices may include, but are not limited to, one or more of a physical keyboard, function keys (such as a volume control key, a switch key, etc.), a trackball, a mouse, a joystick, and the like.

[0242] The display unit 604 can be used to display information input by the user or information provided to the user and various graphical user interfaces of the server, which can be composed of graphics, text, icons, videos and any combination thereof. The display unit 604 may include a display panel. Optionally, the display panel can be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), etc. Further, the touch-sensitive surface can cover the display panel. When the touch-sensitive surface detects a touch operation on or near it, it is transmitted to the processor 608 to determine the type of touch event. The processor 608 then provides corresponding visual output on the display panel according to the type of touch event. Although in Figure 11 In the embodiment, the touch-sensitive surface and the display panel are used as two independent components to realize input and output functions, but in some embodiments, the touch-sensitive surface and the display panel can be integrated to realize input and output functions.

[0243] The server may also include at least one sensor 605, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor may adjust the brightness of the display panel according to the brightness of the ambient light, and the proximity sensor may turn off the display panel and / or backlight when the server is moved to the ear. As a type of motion sensor, the gravity acceleration sensor can detect the magnitude of acceleration in all directions (generally three axes), and can detect the magnitude and direction of gravity when stationary. It can be used for applications that recognize the posture of the mobile phone (such as horizontal and vertical screen switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tapping), etc.; as for other sensors that can be configured on the server, such as gyroscopes, barometers, hygrometers, thermometers, infrared sensors, etc., they will not be described here.

[0244] Audio circuit 606, speakers, and microphones provide an audio interface between the user and the server. Audio circuit 606 converts received audio data into electrical signals and transmits them to the speaker, which then converts them into sound signals for output. Conversely, the microphone converts collected sound signals into electrical signals, which are then received by audio circuit 606 and converted into audio data. The audio data is then processed by output processor 608 and transmitted via RF circuit 601 to, for example, another server. Alternatively, the audio data can be output to memory 602 for further processing. Audio circuit 606 may also include an earphone jack to allow external headphones to communicate with the server.

[0245] WiFi is a short-range wireless transmission technology. The server can help users send and receive emails, browse web pages and access streaming media through the WiFi module 607. It provides users with wireless broadband Internet access. Figure 11 A WiFi module 607 is shown, but it is understandable that it is not an essential component of the server and can be omitted as needed without changing the essence of the invention.

[0246] Processor 608 is the server's control center, connecting various components of the mobile phone using various interfaces and circuits. By running or executing software programs and / or modules stored in memory 602 and accessing data stored in memory 602, it performs various server functions and processes data, thereby providing overall monitoring of the mobile phone. Optionally, processor 608 may include one or more processing cores; preferably, processor 608 may integrate an application processor and a modem processor, with the application processor primarily processing the operating system, user interface, and application programs, while the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into processor 608.

[0247] The server also includes a power supply 609 (e.g., a battery) for supplying power to various components. Preferably, the power supply can be logically connected to the processor 608 via a power management system, thereby enabling the power management system to manage charging, discharging, and power consumption. The power supply 609 can also include one or more DC or AC power supplies, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, and other arbitrary components.

[0248] Although not shown, the server may also include a camera, a Bluetooth module, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 608 in the server will load the executable files corresponding to one or more application processes into the memory 602 according to the following instructions, and the processor 608 will run the application stored in the memory 602 to implement various functions:

[0249] Obtain multiple comments on published content and user interaction information for each comment, including the comment content and comment time;

[0250] Determine the popularity value of each comment information based on the comment information, the user interaction information and the published content;

[0251] Determine the time decay value of each comment information according to the comment time;

[0252] The plurality of comment pieces of information are sorted according to the heat value and the time decay value.

[0253] The server can achieve the effective effects that can be achieved by any of the comment information sorting devices provided in the embodiments of the present application. Please refer to the previous embodiments for details and will not be repeated here.

[0254] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, which may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0255] The above is a detailed introduction to a comment information sorting method, device, storage medium and server provided in an embodiment of 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, based on the ideas of the present invention, there will be changes in the specific implementation methods 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 sorting comment information, characterized in that: include: Obtaining multiple comments on published content and user interaction information for each comment, wherein the comment information includes the comment content and comment time; Determining content features of each comment information based on the comment content and the published content, wherein the content features are determined based on relevance, the number of entity words, and character length, wherein the relevance represents a topic vector distribution distance between the comment content and the published content, and the relevance, the number of entity words, and the character length are respectively positively correlated with the popularity value; Determining the interactive features of each comment information according to the user interactive information; Performing logarithmic processing on the user characteristics, the interaction characteristics, and the content characteristics respectively; Normalizing the user features, the interaction features, and the content features after logarithmic processing using a minimax method to obtain normalized values; According to a preset weighted value, a weighted summation process is performed on the normalized numerical value corresponding to each comment information to obtain a corresponding popularity value, wherein the user characteristics include a like weight value, the like weight value is positively correlated with the popularity value and the user level of the user to whom the comment information belongs, and the user level is negatively correlated with the total number of historical likes, which is obtained by recording the user's like behavior within a historical time period; the user level is obtained by sorting the total number of historical likes and equally dividing the total number of network likes according to the sorting order; Determine a time decay value of each comment information according to the comment time; Sorting the plurality of comment information according to the popularity value and the time decay value; Determine a low-quality classification label for each piece of comment information using a preset classification model; The positions of the sorted plurality of comment information are adjusted according to the low-quality classification labels, and different adjustment ranges are set corresponding to different low-quality classification labels.

2. The comment information sorting method according to claim 1, characterized in that: The determining of the content characteristics of each comment information based on the comment content and the published content includes: Determining the relevance between each of the review contents and the published content; Determining the number of entity words in each of the review contents and the character length of each of the review contents; The relevance, the number of entity words and the character length are used as content features of the corresponding comment information.

3. The comment information sorting method according to claim 2, characterized in that: Determining the relevance between each comment and the published content includes: Determine a first topic vector corresponding to each comment content and a second topic vector corresponding to the published content based on a preset topic model; A distribution distance between the second topic vector and each of the first topic vectors is determined to obtain a relevance between the corresponding comment content and the published content.

4. The comment information sorting method according to claim 1, characterized in that: The user interaction information includes the number of likes for the comment information, and the number of replies and likes elicited by the comment information. Determining the interactive features of each comment information based on the user interaction information includes: Determining the complexity of the comment tree of the comment information according to the number of replies elicited by the comment information; Determining the number of likes in the comment tree of the comment information according to the number of likes induced by the comment information; The number of likes for the comment information, the complexity of the comment tree, and the number of likes for the comment tree are used as interactive features of the corresponding comment information.

5. The comment information sorting method according to claim 1, characterized in that: The comment information also includes the commenting user. Before determining the popularity value of each comment information according to the interactive feature and the content feature, the method further includes: determining historical interaction information of each network user relative to other network users in a network user set, wherein the network user set includes the commenting user; Determining user characteristics of each commenting user based on the historical interaction information; Determining the popularity value of each piece of the comment information according to the interaction feature and the content feature includes: determining the popularity value of each piece of the comment information according to the user feature, the interaction feature and the content feature.

6. The comment information sorting method according to claim 5, characterized in that: The historical interaction information includes the total number of historical likes. The determining of the user characteristics of each commenting user based on the historical interaction information includes: Accumulate the total number of likes in history corresponding to each network user to obtain the total number of likes on the network; Sorting the network users according to the total number of likes in history corresponding to each network user; Determining the user level of each commenting user based on the total number of likes on the network and the ranked network users; According to the user level of each commenting user, a corresponding like weight value is determined, and the like weight value is used as a user feature.

7. The comment information sorting method according to claim 5, characterized in that: The historical interaction information includes historical likes or replies, and determining the user characteristics of each commenting user based on the historical interaction information includes: Constructing a network node graph using the historical likes or reply relationships as edges and the network users as nodes; According to a preset user rating algorithm and the network node graph, a node weight value of each commenting user is determined, and the node weight value is used as a user feature.

8. The comment information sorting method according to any one of claims 1 to 7, characterized in that: The sorting of the plurality of comment information according to the popularity value and the time decay value includes: Calculate the product of the popularity value and the time decay value corresponding to each comment information to obtain the recommendation degree; The plurality of review information are sorted according to the numerical values of the recommendation degrees.

9. The comment information sorting method according to any one of claims 1 to 7, characterized in that: After sorting the plurality of comment information according to the popularity value and the time decay value, the method further includes: displaying the adjusted plurality of comment information on the comment interface of the published content.

10. A comment information sorting device, characterized in that: include: An acquisition module, configured to acquire multiple comments on published content and user interaction information for each comment, wherein the comment information includes the comment content and comment time; A first determination module is configured to determine content features of each of the comment information based on the comment content and the published content, wherein the content features are determined based on relevance, the number of entity words, and character length, wherein the relevance represents the similarity between the comment content and the published content, and the relevance, the number of entity words, and the character length are respectively positively correlated with the popularity value; determine interactive features of each of the comment information based on the user interactive information; and perform logarithmic processing on the user features, the interactive features, and the content features; Normalizing the user features, the interaction features, and the content features after logarithmic processing using a minimax method to obtain normalized values; According to a preset weighted value, a weighted summation process is performed on the normalized numerical value corresponding to each comment information to obtain a corresponding popularity value, wherein the user characteristics include a like weight value, the like weight value is positively correlated with the popularity value and the user level of the user to whom the comment information belongs, and the user level is negatively correlated with the total number of historical likes, which is obtained by recording the user's like behavior within a historical time period; the user level is obtained by sorting the total number of historical likes and equally dividing the total number of network likes according to the sorting order; A second determining module is configured to determine a time decay value of each comment information according to the comment time; A sorting module, configured to sort the plurality of comment information according to the popularity value and the time decay value; Adjusting the display module to determine a low-quality classification label for each review information using a preset classification model; The positions of the sorted plurality of comment information are adjusted according to the low-quality classification labels, and different adjustment ranges are set corresponding to different low-quality classification labels.

11. The comment information sorting device according to claim 10, characterized in that: The first determining module includes a first determining unit configured to: Determining the relevance between each of the review contents and the published content; Determining the number of entity words in each of the review contents and the character length of each of the review contents; The relevance, the number of entity words and the character length are used as content features of the corresponding comment information.

12. The comment information sorting device according to claim 11, characterized in that: The first determining unit is specifically configured to: Determine a first topic vector corresponding to each comment content and a second topic vector corresponding to the published content based on a preset topic model; A distribution distance between the second topic vector and each of the first topic vectors is determined to obtain a relevance between the corresponding comment content and the published content.

13. The comment information sorting device according to claim 10, characterized in that: The user interaction information includes the number of likes for the comment information, and the number of replies and likes elicited by the comment information. The first determining module includes a second determining unit configured to: Determining the complexity of the comment tree of the comment information according to the number of replies elicited by the comment information; Determining the number of likes in the comment tree of the comment information according to the number of likes induced by the comment information; The number of likes for the comment information, the complexity of the comment tree, and the number of likes for the comment tree are used as interactive features of the corresponding comment information.

14. The comment information sorting device according to claim 10, characterized in that: The comment information also includes the commenting user, and the first determining module includes a third determining unit and a fourth determining unit, wherein the fourth determining unit is configured to: Before determining the popularity value of each comment information according to the interaction feature and the content feature, determining historical interaction information of each network user relative to other network users in a network user set, wherein the network user set includes the commenting user; Determining user characteristics of each commenting user based on the historical interaction information; The third determining unit is configured to determine the popularity value of each piece of comment information according to the user characteristics, the interaction characteristics, and the content characteristics.

15. The comment information sorting device according to claim 14, characterized in that: The historical interaction information includes the total number of historical likes, and the fourth determining unit is specifically configured to: Accumulate the total number of likes in history corresponding to each network user to obtain the total number of likes on the network; Sorting the network users according to the total number of likes in history corresponding to each network user; Determining the user level of each commenting user based on the total number of likes on the network and the ranked network users; According to the user level of each commenting user, a corresponding like weight value is determined, and the like weight value is used as a user feature.

16. The comment information sorting device according to claim 14, characterized in that: The historical interaction information includes historical likes or replies, and the fourth determining unit is specifically configured to: Constructing a network node graph using the historical likes or reply relationships as edges and the network users as nodes; According to a preset user rating algorithm and the network node graph, a node weight value of each commenting user is determined, and the node weight value is used as a user feature.

17. The comment information sorting device according to any one of claims 10 to 16, characterized in that: The sorting module is specifically used for: Calculate the product of the popularity value and the time decay value corresponding to each comment information to obtain the recommendation degree; The plurality of review information are sorted according to the numerical values of the recommendation degrees.

18. The comment information sorting device according to any one of claims 10 to 16, characterized in that: Also included is an adjustment display module for: After sorting the plurality of review information according to the popularity value and the time decay value, determining a low-quality classification label for each review information using a preset classification model; Adjusting the positions of the sorted plurality of review information according to the low-quality classification labels; The adjusted plurality of comment information is displayed on the comment interface of the published content.

19. A computer-readable storage medium, characterized in that The storage medium stores a plurality of instructions, which are suitable for being loaded by a processor to execute the comment information sorting method according to any one of claims 1 to 9.

20. A server, characterized in that: It includes a processor and a memory, the processor is electrically connected to the memory, the memory is used to store instructions and data, and the processor is used to execute the steps in the comment information sorting method according to any one of claims 1 to 9.

Images