Comment sorting method and apparatus

Abstract

This disclosure provides a comment ranking method and apparatus, relating to the field of artificial intelligence technology, particularly to the fields of natural language processing, computer vision, and deep learning. A specific implementation of the method includes: extracting features from internal comments of a target video published in an internal application to generate internal comment features; extracting features from external comments of a target video published in an external application to generate external comment features; inputting the internal comment features into an internal modality optimization model to output an internal ecological value score; and inputting the external comment features into an external empirical ranking model to output an external empirical ranking score; fusing the internal ecological value score and the external empirical ranking score to generate a ranking score for each internal comment; and ranking each internal comment based on its ranking score.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, particularly to the fields of natural language processing, computer vision, and deep learning. Background Technology

[0002] With the rapid development of internet content platforms such as short video and text-based communities, the scale of UGC (User Generated Content) continues to expand. Comments, as a crucial medium for user interaction and content dissemination, are ranked in a way that directly impacts user browsing experience and community ecosystem development. Given the massive amounts of comment data, how to efficiently and accurately present high-value content has become a critical direction for platform information processing.

[0003] Current mainstream comment ranking systems typically employ a three-tiered architecture: coarse ranking, fine ranking, and manual intervention. The coarse ranking stage uses pre-defined semantic rules and formulas, combined with factors such as likes, replies, and posting time decay, to initially filter all comments and select the Top N candidate comments. The fine ranking stage inputs the coarse ranking results into a shallow machine learning model, using basic interaction features of users, authors, and comments to complete the scoring. Finally, operations personnel manually adjust the ranking based on configured rules. This ranking logic is often deeply coupled with comment database queries, index building, and front-end pagination logic. Summary of the Invention

[0004] This disclosure provides a comment sorting method, apparatus, device, storage medium, and program product.

[0005] In a first aspect, embodiments of this disclosure propose a comment ranking method, comprising: extracting features from internal comments of a target video published in an internal application to generate internal comment features, and extracting features from external comments of a target video published in an external application to generate external comment features; inputting the internal comment features into an internal modality optimization model to output an internal ecological value score, and inputting the external comment features into an external empirical ranking model to output an external empirical ranking score; fusing the internal ecological value score and the external empirical ranking score to generate a ranking score for each internal comment; and ranking each internal comment based on its ranking score.

[0006] Secondly, this disclosure proposes an internal modality optimization model training method, comprising: extracting features from sample internal comments of sample videos published in an internal application to generate sample internal comment features; constructing sample partial order pairs based on the sample internal comment features, wherein the sample partial order pairs include the posterior value score and information value score of the sample internal comments; using the sample internal comment features as input and the sample partial order pairs as training labels to train a ranking model to learn the partial order relationship, thereby obtaining an internal modality optimization model.

[0007] Thirdly, this disclosure proposes an external empirical ordering model training method, including: extracting features from sample external comments of sample videos published in external applications to generate sample external comment features; determining the sample external empirical ordering score of the sample comments based on the sample external comment features; using the sample external comment features as input and the sample external empirical ordering score as training labels to train the ordering model to learn relative ordering relationships and / or absolute ordering positions, thereby obtaining the external empirical ordering model.

[0008] Fourthly, embodiments of this disclosure propose a comment ranking apparatus, comprising: an extraction module configured to extract features from internal comments of a target video published in an internal application to generate internal comment features, and to extract features from external comments of a target video published in an external application to generate external comment features; an output module configured to input the internal comment features into an internal modality optimization model to output an internal ecological value score, and to input the external comment features into an external empirical ranking model to output an external empirical ranking score; a fusion module configured to fuse the internal ecological value score and the external empirical ranking score to generate a ranking score for each internal comment; and a ranking module configured to rank each internal comment based on its ranking score.

[0009] Fifthly, embodiments of this disclosure propose an internal modality optimization model training apparatus, comprising: an extraction module configured to extract features from sample internal comments of sample videos published in an internal application to generate sample internal comment features; a construction module configured to construct sample partial order pairs based on the sample internal comment features, wherein the sample partial order pairs include posterior value scores and information value scores of the sample internal comments; and a training module configured to use the sample internal comment features as input and the sample partial order pairs as training labels to train a ranking model to learn the partial order relationship and obtain an internal modality optimization model.

[0010] In a sixth aspect, embodiments of this disclosure propose an external empirical ordering model training apparatus, comprising: an extraction module configured to extract features from sample external comments of sample videos published in an external application to generate sample external comment features; a determination module configured to determine the sample external empirical ordering score of the sample comments based on the sample external comment features; and a training module configured to take the sample external comment features as input, take the sample external empirical ordering score as training labels, train the ordering model to learn relative ordering relationships and / or absolute ordering positions, and obtain an external empirical ordering model.

[0011] In a seventh aspect, embodiments of this disclosure provide an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the methods described in the first, second, or third aspect.

[0012] Eighthly, embodiments of this disclosure provide a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the methods described in the first, second, or third aspect.

[0013] In a ninth aspect, embodiments of this disclosure provide a computer program product including a computer program that, when executed by a processor, implements the methods described in the first, second, or third aspect.

[0014] The key or essential features of the embodiments disclosed herein are not intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0015] Other features, objects, and advantages of this disclosure will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are provided for a better understanding of the invention and are not intended to limit the scope of this disclosure. Wherein: Figure 1 This is a flowchart of an embodiment of the internal modality optimization model training method according to the present disclosure; Figure 2 This is the network architecture diagram of the internal modal optimization model; Figure 3 This is a flowchart of an embodiment of the external empirical sequence model training method according to this disclosure; Figure 4 This is a network architecture diagram of an external empirical sequence model; Figure 5 This is a flowchart of an embodiment of the comment sorting method according to this disclosure; Figure 6 This is the architecture diagram of the comment sorting service; Figure 7 This is a schematic diagram of the structure of an embodiment of the internal modality optimization model training device according to the present disclosure; Figure 8 This is a schematic diagram of the structure of an embodiment of an external empirical sequence model training device according to the present disclosure; Figure 9 This is a schematic diagram of the structure of one embodiment of the comment sorting device according to the present disclosure; Figure 10This is a block diagram of an electronic device used to implement the comment sorting method of the embodiments of this disclosure. Detailed Implementation

[0016] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other. This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] Figure 1 A flowchart 100 is shown as an embodiment of an internal modality optimization model training method according to the present disclosure. The internal modality optimization model training method includes the following steps: Step 101: Extract features from the sample internal comments of the sample videos published in the internal application to generate sample internal comment features.

[0019] In this embodiment, the entity executing the internal modality optimization model training method can extract features from the sample internal comments of sample videos published in the internal application to generate sample internal comment features.

[0020] Internal comment features may include, but are not limited to: modal features, posterior interaction features, author features, and contextual features. Modal features may include, but are not limited to: whether the comment contains text, whether it contains images, whether it contains emojis, text length, and text sentiment. Posterior interaction features may include, but are not limited to: number of likes, number of replies, number of shares, and click-through rate. Author features may include, but are not limited to: author account attributes and interaction history. Contextual features may include, but are not limited to: the popularity of the sample video and its publication time.

[0021] Step 102: Based on the intra-sample comment features, construct sample partial order pairs, which include the posterior value score and information value score of the intra-sample comments.

[0022] In this embodiment, the aforementioned execution entity can construct sample partial order pairs based on the intra-sample comment features. The sample partial order pairs include the posterior value score and information value score of the intra-sample comments.

[0023] The posterior score can be calculated using a predefined piecewise function based on the sum of likes and replies to the sample comments. The information value score can be calculated using a weighted average of the rich media modalities of the sample comments, for example, a weighted sum of text weight 1.5, image weight 1, and emoji weight 0.5. The constructed partial order pairs can include: comments with high posterior and high information value paired with comments with high posterior and low information value, and comments with high posterior and low information value paired with comments with low posterior and low information value.

[0024] Step 103: Using the internal comment features of the sample as input and the partial order pairs of the sample as training labels, train the ranking model to learn the partial order relationship and obtain the internal modality optimization model.

[0025] In this embodiment, the aforementioned execution entity can use the internal comment features of the sample as input, use the sample partial order pairs as training labels, train the ranking model to learn the partial order relationship, and obtain the internal modality optimization model.

[0026] The model can employ a pairwise ranking learning model, with a simple MLP (Multilayer Perceptron) neural network at the bottom layer. The model's learning objective is to fit the partial order relationship between reviews, and finally output a score to represent the ecological value of the platform.

[0027] This disclosure provides an internal modality optimization model training method. By extracting multi-dimensional features of sample comments, a partial order pair based on posterior values ​​and information value is constructed, and a model that can accurately quantify the ecological value of internal comments is trained, providing a reliable internal scoring basis for subsequent comment ranking.

[0028] Figure 2 The network architecture diagram of the internal modality optimization model is shown. This model takes multi-dimensional features of comments as input, processes them through an MLP neural network and activation functions, and outputs an internal ecosystem value score. The input layer contains three types of features: the first type is posterior features of the comment, including likes, replies, shares, and click-through rate; the second type is basic features of the comment, including whether it contains censored words (low quality score), text length, character ratio, whether the author liked it, and whether the author replied; the third type is modal features of the comment, including whether it contains text, emojis, and images. These features are concatenated and input into the MLP, then normalized using a sigmoid activation function, finally outputting a normalized internal ecosystem value score, which is the normalized result of fusing the information value score and the posterior value score.

[0029] Figure 3 A flow 300 of an embodiment of an external empirical ordering model training method according to the present disclosure is shown. This external empirical ordering model training method includes the following steps: Step 301: Extract features from the sample external comments of the sample videos published in the external application to generate sample external comment features.

[0030] In this embodiment, the execution entity of the external empirical sequence model training method can extract features from sample external comments of sample videos published in external applications to generate sample external comment features.

[0031] External and internal comment features are partially shared, including but not limited to: basic comment features, modal features, and posterior interaction features, to ensure consistency and transferability of features across platforms.

[0032] Step 302: Based on the external comment features of the sample, determine the external empirical ranking score of the sample comments.

[0033] In this embodiment, the aforementioned execution entity can determine the external empirical ranking score of the sample comments based on the external comment features.

[0034] The external empirical ranking score can be determined based on the ranking results of real comments on external platforms for the same author and the same video, including relative ranking relationships and absolute ranking positions, and is used to characterize the ranking preferences of external communities for comment quality.

[0035] Step 303: Using the external comment features of the sample as input and the external empirical order ranking score of the sample as training label, train the ranking model to learn the relative ranking relationship and / or absolute ranking position, and obtain the external empirical order model.

[0036] In this embodiment, the aforementioned execution entity can use the external comment features of the sample as input, use the external empirical order ranking score of the sample as training label, train the ranking model to learn the relative ranking relationship and / or absolute ranking position, and obtain the external empirical order model.

[0037] The model structure is consistent with the internal modal optimization model. It can use a simple MLP neural network, with only the training labels and learning objectives being different. The model outputs a score that represents the ranking tendency of the external platform, realizing the transfer and absorption of advanced external ranking experience.

[0038] This disclosure provides an external empirical ranking model training method. By training the model with real-world ranking supervision signals of comments from the same author and video across platforms, the model is equipped with the ability to learn the ranking rules of high-quality external communities, thus introducing advanced industry experience into comment ranking.

[0039] Figure 4The network architecture diagram of the external experience ranking model is shown. The model processes external comments from external resources, and its input features include basic comment features, modal features, and posterior features. These features are uniformly input into an MLP network, using the rank ranking results of external platform comments as the learning objective. The model outputs an external experience ranking tendency score after passing through a sigmoid activation function. By learning the relative ranking relationships and absolute ranking positions of external comments, the model transfers the ranking experience of mature external communities to the internal platform, achieving a fusion of internal and external ranking advantages.

[0040] Figure 5 A flow 500 is shown as an embodiment of a comment sorting method according to the present disclosure. The comment sorting method includes the following steps: Step 501: Extract features from internal comments of the target video published in the internal application to generate internal comment features, and extract features from external comments of the target video published in the external application to generate external comment features.

[0041] In this embodiment, the entity executing the comment ranking method can extract features from internal comments of target videos published in internal applications to generate internal comment features, and extract features from external comments of target videos published in external applications to generate external comment features.

[0042] The execution entity for the comment ranking method can be an independently deployed comment ranking microservice. This microservice is decoupled from the comment storage service, indexing service, and pagination loading service, and communicates with the user through a standardized API (Application Programming Interface). When a comment display request is received from a client, the execution entity first retrieves all internal comments under the target video published by the same author within the internal application, and then performs multi-dimensional feature extraction on each internal comment to generate internal comment features. Internal comment features may include, but are not limited to: comment modal features, used to characterize the rich media information of the comment, including whether it contains text, images, emoticons, the number of emoticons, text length, text sentiment, and whether it contains low-quality censored words; comment post-interaction features, used to characterize user interaction with the comment, including likes, replies, shares, click-through rate, and interaction timeliness; comment basic features, including comment publication time, character percentage, and content quality score; author features, including author account attributes, historical interaction behavior, and whether the author liked or replied to the current comment; and contextual features, including the target video's popularity level, video category, publication duration, and content pool.

[0043] Meanwhile, the aforementioned implementing entities obtain external comments on the same target video published by the same author in external applications, extract them according to the feature system consistent with the internal comment features, and generate external comment features. The external comment features and internal comment features are partially shared, including but not limited to: basic comment features, comment modal features, and comment post-interaction features, to ensure the consistency of cross-platform model input and facilitate the effective transfer of external ranking experience.

[0044] Step 502: Input the internal comment features into the internal modality optimization model and output the internal ecological value score; input the external comment features into the external empirical ranking model and output the external empirical ranking score.

[0045] In this embodiment, the aforementioned executing entity can input internal comment features into an internal modality optimization model to output an internal ecological value score, and input external comment features into an external empirical ranking model to output an external empirical ranking score.

[0046] Internal review features are input into a trained internal modality optimization model. The model performs inference calculations based on the input multi-dimensional features and outputs an internal ecosystem value score that characterizes the quality of a review within the platform's internal ecosystem. Simultaneously, external review features are input into a trained external empirical ranking model. The model performs inference calculations based on the input shared features and outputs an external empirical ranking score that characterizes the ranking tendency of reviews on external platforms.

[0047] In some embodiments, the internal modal optimization model calculates the internal ecological value score through the following steps: First, based on the characteristics of internal comments, information value scores and posterior scores of internal comments are generated.

[0048] Specifically, the information value score of an internal comment is generated by weighted summation of the number of various rich media modalities contained within it; the posterior interaction count of the internal comment is then input into a piecewise function to generate the posterior value score of the internal comment. For example, the weight for the text modality is 1.5, the weight for the image modality is 1, and the weight for the emoji modality is 0.5. The posterior interaction count can be the sum of the number of likes and replies to the comment.

[0049] The piecewise function can be defined as follows: When the number of posterior interactions In the interval [1, 10], the piecewise function is the power of 1 / 10 of the posterior interaction number, that is, ; When the number of posterior interactions In the interval [10, 100], the piecewise function is the logarithm of the number of posterior interactions to the base 3, i.e., ; When the number of posterior interactions In the interval [100, 500], the piecewise function is the quotient of the number of posterior interactions divided by 100, plus 4, that is, ; When the number of posterior interactions In the interval [500, 1000], the piecewise function is the quotient of the number of posterior interactions divided by 90, plus 4, that is, ; When the number of posterior interactions When the number is greater than or equal to 1000, the piecewise function is the quotient of the number of posterior interactions divided by 60, minus 1, that is, .

[0050] Then, based on the information value score and posterior value score of the internal comments, the internal ecosystem value score is calculated.

[0051] Specifically, the sum of the information value score and the posterior value score of the internal comments is calculated, and the sum is normalized to generate an internal ecological value score. The sum of the information value score and the posterior value score of the internal comments yields a fusion score, which is then normalized to map to a preset range, generating a stable and comparable internal ecological value score.

[0052] Step 503: Integrate the internal ecological value score and the external experience ranking score to generate a ranking score for each internal comment.

[0053] In this embodiment, the aforementioned executing entity can integrate internal ecological value scores and external experience ranking scores to generate a ranking score for each internal comment.

[0054] By employing a nonlinear weighted fusion method, the internal ecological value score and the external experience ranking score can be fused and calculated.

[0055] In some embodiments, the steps of the nonlinear weighted fusion method can be as follows: First, the internal ecological value score is linearly weighted and subjected to a power-law nonlinear transformation to generate an internal nonlinear transformation score. Similarly, the external empirical ranking score is linearly weighted and subjected to a power-law nonlinear transformation to generate an external nonlinear transformation score. Specifically, linear weighting is achieved through manually tuned hyperparameters, followed by the nonlinear transformation using a power function.

[0056] Then, the product of the internal nonlinear transformation score and the external nonlinear transformation score is calculated to generate a ranking score for each internal comment.

[0057] The specific fusion formula can be: Ranking score = pow ((out_p1 + out_p2 × out_q), out_p3) × pow ((inner_p1 + inner_p2 × inner_q), inner_p3).

[0058] Wherein, out_q is the external experience ranking score, out_p1, out_p2, and out_p3 are the hyperparameters corresponding to the external model, with values ​​ranging from 1 to 5, 1 to 5, and 0 to 2, respectively; inner_q is the internal ecological value score, and inner_p1, inner_p2, and inner_p3 are the hyperparameters corresponding to the internal model, with values ​​ranging from 1 to 5, 1 to 5, and 0 to 2, respectively. The optimal combination of all hyperparameters was determined after verifying their effectiveness through online small-scale experiments.

[0059] Step 504: Sort each internal comment based on its ranking score.

[0060] In this embodiment, the aforementioned executing entity can sort each internal comment based on its ranking score.

[0061] The initial ranking is obtained by sorting the internal comments in descending order of their ranking scores. Further fine-tuning of this initial ranking, based on platform operational strategies, includes, but is not limited to: appropriately boosting the ranking of newly posted comments, downgrading the ranking of high-risk comments, and filtering low-quality content. After fine-tuning, the ranked list of comment IDs is returned to the front-end client. The client then renders and displays the comments in this order, allowing users to prioritize high-quality comments that are information-dense, rich in multimedia, highly interactive, and aligned with both internal ecosystem trends and external popular trends, thereby improving user interaction and the quality of the community ecosystem.

[0062] The comment ranking method provided in this disclosure achieves system decoupling through a microservice architecture, fully explores the value of rich media comments through an internal modal optimization model, introduces advanced external ranking experience through an external empirical ranking model, and integrates internal and external advantages through a nonlinear fusion method, which can significantly improve the accuracy of comment ranking, content richness, and system iteration efficiency.

[0063] Figure 6The diagram illustrates the architecture of the comment ranking service. The overall architecture consists of four layers: basic ranking and filtering, online fine-tuning, multi-objective fusion of comments, and comment output. The bottom layer accesses the full volume of comment data, relying on data sources including basic data, tag data, and posterior data. Initial filtering is completed through offline coarse-grained ranking and comment index construction. In the online fine-tuning stage, posterior features, multimodal semantic features, and sentiment / opinion features are constructed for comments, combined with a large-scale multi-objective ranking model and a GBDT (Gradient Boosting Decision Tree) model for ranking. The middle layer, through a multi-objective fusion module, non-linearly fuses internal ecosystem value scores with external experience ranking scores, and supports operational intervention strategies. Finally, a ranked list of comments is output, with a maximum of 20 comments displayed on the front end. This architecture decouples the comment ranking service as a microservice from the underlying storage, indexing, and pagination services, supporting rapid model iteration and A / B testing.

[0064] Further reference Figure 7 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of an internal modal optimization model training device, which is similar to... Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0065] like Figure 7 As shown, the internal modality optimization model training device 700 of this embodiment may include: an extraction module 701, a construction module 702, and a training module 703. The extraction module 701 is configured to extract features from sample internal comments of sample videos published in the internal application, generating sample internal comment features; the construction module 702 is configured to construct sample partial order pairs based on the sample internal comment features, where each sample partial order pair includes the posterior value score and information value score of the sample internal comment; the training module 703 is configured to use the sample internal comment features as input and the sample partial order pairs as training labels to train a ranking model to learn the partial order relationship, thereby obtaining the internal modality optimization model.

[0066] In this embodiment, the specific processing of the extraction module 701, the construction module 702, and the training module 703 in the internal modal optimization model training device 700, and the resulting technical effects, can be found in the following references: Figure 1 The relevant descriptions of steps 101-103 in the corresponding embodiments will not be repeated here.

[0067] Further reference Figure 8 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of an external empirical sequence model training device, which is similar to... Figure 3 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0068] like Figure 8 As shown, the external empirical ordering model training device 800 of this embodiment may include: an extraction module 801, a determination module 802, and a training module 803. The extraction module 801 is configured to extract features from sample external comments of sample videos published in external applications to generate sample external comment features; the determination module 802 is configured to determine the sample external empirical ordering score of the sample comments based on the sample external comment features; and the training module 803 is configured to use the sample external comment features as input and the sample external empirical ordering score as training labels to train the ordering model to learn relative ordering relationships and / or absolute ordering positions, thereby obtaining the external empirical ordering model.

[0069] In this embodiment, the specific processing of the extraction module 801, the determination module 802, and the training module 803 in the external empirical sequence model training device 800, and the resulting technical effects, can be found in reference to [reference needed]. Figure 3 The relevant descriptions of steps 301-303 in the corresponding embodiments will not be repeated here.

[0070] Further reference Figure 9 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of a comment sorting device, which is similar to... Figure 5 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0071] like Figure 9 As shown, the comment ranking device 900 of this embodiment may include: an extraction module 901, an output module 902, a fusion module 903, and a ranking module 904. The extraction module 901 is configured to extract features from internal comments of target videos published in internal applications to generate internal comment features, and to extract features from external comments of target videos published in external applications to generate external comment features. The output module 902 is configured to input internal comment features into an internal modality optimization model to output an internal ecological value score, and to input external comment features into an external empirical ranking model to output an external empirical ranking score. The fusion module 903 is configured to fuse the internal ecological value score and the external empirical ranking score to generate a ranking score for each internal comment. The ranking module 904 is configured to rank each internal comment based on its ranking score.

[0072] In this embodiment, the specific processing of the extraction module 901, output module 902, fusion module 903, and sorting module 904 in the comment sorting device 900, and the resulting technical effects, can be found in reference to [reference needed]. Figure 5 The relevant descriptions of steps 501-504 in the corresponding embodiments will not be repeated here.

[0073] In some optional implementations of this embodiment, the output module 902 is further configured to: generate information value scores and posterior scores for internal comments based on internal comment features; and calculate internal ecological value scores based on the information value scores and posterior scores for internal comments.

[0074] In some optional implementations of this embodiment, the output module 902 is further configured to: perform a weighted summation of the number of various rich media modalities contained in the internal comment to generate an information value score for the internal comment; and input the posterior interaction number of the internal comment into a piecewise function to generate a posterior value score for the internal comment.

[0075] In some optional implementations of this embodiment, the piecewise function is defined as follows: when the number of posterior interactions is in the interval [1, 10], the piecewise function is the power of 1 / 10 of the number of posterior interactions; when the number of posterior interactions is in the interval [10, 100], the piecewise function is the logarithm of the number of posterior interactions to the base 3; when the number of posterior interactions is in the interval [100, 500], the piecewise function is the quotient of the number of posterior interactions divided by 100, plus 4; when the number of posterior interactions is in the interval [500, 1000], the piecewise function is the quotient of the number of posterior interactions divided by 90, plus 4; when the number of posterior interactions is greater than or equal to 1000, the piecewise function is the quotient of the number of posterior interactions divided by 60, minus 1.

[0076] In some optional implementations of this embodiment, the output module 902 is further configured to: calculate the sum of the information value score of the internal comment and the posterior value score of the internal comment, and normalize the sum to generate an internal ecological value score.

[0077] In some optional implementations of this embodiment, the fusion module 903 is further configured to: perform linear weighting and power-law nonlinear transformation on the internal ecological value score to generate an internal nonlinear transformation score, and perform linear weighting and power-law nonlinear transformation on the external empirical ranking score to generate an external nonlinear transformation score; calculate the product of the internal nonlinear transformation score and the external nonlinear transformation score to generate a ranking score for each internal comment.

[0078] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0079] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0080] Figure 10A schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0081] like Figure 10 As shown, device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 1002 or a computer program loaded into random access memory (RAM) 1003 from storage unit 1008. The RAM 1003 may also store various programs and data required for the operation of device 1000. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via bus 1004. Input / output (I / O) interface 1005 is also connected to bus 1004.

[0082] Multiple components in device 1000 are connected to I / O interface 1005, including: input unit 1006, such as keyboard, mouse, etc.; output unit 1007, such as various types of monitors, speakers, etc.; storage unit 1008, such as disk, optical disk, etc.; and communication unit 1009, such as network card, modem, wireless transceiver, etc. Communication unit 1009 allows device 1000 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0083] The computing unit 1001 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as the comment ranking method. For example, in some embodiments, the comment ranking method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed on device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by the computing unit 1001, one or more steps of the comment ranking method described above may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to perform the comment ranking method by any other suitable means (e.g., by means of firmware).

[0084] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0085] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0086] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0087] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0088] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0089] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, distributed system servers, or servers incorporating blockchain technology.

[0090] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution provided in this disclosure can be achieved, and this is not limited herein.

[0091] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for sorting comments, comprising: Feature extraction is performed on internal comments of the target video published in an internal application to generate internal comment features, and feature extraction is performed on external comments of the target video published in an external application to generate external comment features; The internal comment features are input into the internal modality optimization model to output the internal ecological value score, and the external comment features are input into the external empirical ranking model to output the external empirical ranking score. By combining the internal ecological value score and the external experience ranking score, a ranking score is generated for each internal comment; Each internal comment is ranked based on its ranking score.

2. The method according to claim 1, wherein, The step of inputting the internal comment features into the internal modality optimization model and outputting the internal ecosystem value score includes: Based on the internal comment features, generate the information value score and posterior value score of the internal comments; The internal ecological value score is calculated based on the information value score and posterior value score of the internal comments.

3. The method according to claim 2, wherein, The process of generating the information value score and posterior value score of the internal comment based on the internal comment features includes: The information value score of the internal comment is generated by weighted summing of the number of various rich media modalities contained in the internal comment. The posterior interaction count of the internal comments is input into the piecewise function to generate the posterior score of the internal comments.

4. The method according to claim 3, wherein, The piecewise function is defined as follows: When the number of posterior interactions is in the interval [1, 10], the piecewise function is the power of 1 / 10 of the number of posterior interactions; When the number of posterior interactions is in the interval [10, 100], the piecewise function is the logarithm of the number of posterior interactions to the base 3; When the number of posterior interactions is in the interval [100, 500], the piecewise function is the quotient of the number of posterior interactions divided by 100, plus 4; When the number of posterior interactions is in the interval [500, 1000], the piecewise function is the quotient of the number of posterior interactions divided by 90, plus 4; When the number of posterior interactions is greater than or equal to 1000, the piecewise function is the quotient of the number of posterior interactions divided by 60, minus 1.

5. The method according to claim 2, wherein, The calculation of the internal ecological value score based on the information value score and posterior value score of the internal comments includes: The information value score of the internal comment is calculated as the sum of the posterior value score of the internal comment, and the sum is normalized to generate the internal ecological value score.

6. The method according to claim 1, wherein, The process of integrating the internal ecological value score and the external experience ranking score to generate a ranking score for each internal comment includes: The internal ecological value score is subjected to linear weighting and power-law nonlinear transformation to generate an internal nonlinear transformation score, and the external empirical ranking score is subjected to linear weighting and power-law nonlinear transformation to generate an external nonlinear transformation score. The product of the internal nonlinear transformation score and the external nonlinear transformation score is calculated to generate a ranking score for each internal comment.

7. A method for training an internal modality optimization model, comprising: Extract features from the internal comments of sample videos published in the internal application to generate sample internal comment features; Based on the intra-sample comment features, sample partial order pairs are constructed, wherein the sample partial order pairs include the posterior value score and information value score of the intra-sample comments; Using the internal comment features of the sample as input and the partial order pairs of the sample as training labels, the ranking model is trained to learn the partial order relationship, thus obtaining the internal modality optimization model.

8. A method for training an external empirical sequence model, comprising: Extract features from external comments on sample videos published in external applications to generate sample external comment features; Based on the external comment features of the sample, determine the external empirical ranking score of the sample comments; Using the external comment features of the sample as input and the external empirical order ranking score of the sample as training labels, the ranking model is trained to learn the relative ranking relationship and / or absolute ranking position, thus obtaining the external empirical order model.

9. A comment sorting device, comprising: The extraction module is configured to extract features from internal comments of a target video published in an internal application to generate internal comment features, and to extract features from external comments of the target video published in an external application to generate external comment features. The output module is configured to input the internal comment features into an internal modality optimization model and output an internal ecological value score, and to input the external comment features into an external empirical ranking model and output an external empirical ranking score. The fusion module is configured to fuse the internal ecological value score and the external experience ranking score to generate a ranking score for each internal comment; The sorting module is configured to sort each internal comment based on its sort score.

10. An internal modal optimization model training device, comprising: The extraction module is configured to extract features from sample internal comments of sample videos published in the internal application and generate sample internal comment features; The construction module is configured to construct sample partial order pairs based on the intra-sample comment features, wherein the sample partial order pairs include the posterior value score and information value score of the intra-sample comments; The training module is configured to take the internal comment features of the sample as input, use the partial order pairs of the sample as training labels, train the ranking model to learn the partial order relationship, and obtain the internal modality optimization model.

11. An external empirical sequence model training device, comprising: The extraction module is configured to extract features from sample external comments of sample videos published in external applications and generate sample external comment features. The determination module is configured to determine the external empirical ranking score of the sample comments based on the external comment features of the sample; The training module is configured to take the external comment features of the sample as input, the external empirical order ranking score of the sample as training label, and train the ranking model to learn the relative ranking relationship and / or absolute ranking position to obtain the external empirical order model.

12. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6, 7, or 8.

13. A non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the method of any one of claims 1-6, 7, or 8.

14. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-6, 7, or 8.

Images