A title generation method, device, apparatus and storage medium

Abstract

The present disclosure relates to a title generation method and device, equipment and storage medium, and belongs to the technical field of Internet, and can solve the problem of low efficiency of general title generation method. The title generation method comprises: obtaining target resource data of a target multimedia resource; the target resource data comprises an initial title and resource content of the target multimedia resource; inputting the target resource data into a pre-trained title generation model to obtain a first push title of the target multimedia resource; the title generation model is trained according to first sample resource data of a first sample multimedia resource; the first sample resource data comprises an initial title, a push title and resource content of the first sample multimedia resource.

Description

Technical Field

[0001] This disclosure relates to the field of Internet technology, and in particular to a title generation method, apparatus, device, and storage medium. Background Technology

[0002] To increase the click-through rate of multimedia resources, it is necessary to generate push titles that can quickly capture users' interest before pushing multimedia resources.

[0003] The common method is to generate push titles for multimedia resources through manual editing, combined with subjective judgment and understanding. However, when the workload is large, manual editing is inefficient. Summary of the Invention

[0004] This disclosure provides a title generation method, apparatus, device, and storage medium to solve the problem of low efficiency in general title generation methods.

[0005] The technical solution of this disclosure is as follows:

[0006] According to a first aspect of the present disclosure, a title generation method is provided, comprising: acquiring target resource data of a target multimedia resource; the target resource data including an initial title and resource content of the target multimedia resource; inputting the target resource data into a pre-trained title generation model to obtain a first push title of the target multimedia resource; the title generation model is trained based on first sample resource data of a first sample multimedia resource; the first sample resource data including: the initial title, push title, and resource content of the first sample multimedia resource.

[0007] Optionally, the title generation method further includes: inputting the push titles of the first sample multimedia resources into a pre-trained title classification model to obtain a first push title set and a second push title set; the click-through rate of the push titles in the first push title set is greater than the click-through rate of the push titles in the second push title set; performing a pairing operation on the first push title set and the second push title set according to a relevance algorithm to obtain multiple push title pairs; a push title pair includes: one push title in the first push title set, and one push title in the second push title set whose relevance to one push title in the first push title set is greater than a preset relevance; using the push titles in the first push title set among the multiple push title pairs as the training target, training a first preset model based on the push titles in the second push title set among the multiple push title pairs and the first resource content, until the first preset model meets the model convergence condition to obtain a first title generation model; the first resource content includes: the resource content of the first sample multimedia resources corresponding to the push titles in the second push title set among the multiple push title pairs; and determining the title generation model according to the first title generation model.

[0008] Optionally, the method for determining the title generation model based on the first title generation model includes: determining the first title generation model as the title generation model; or, using the push titles in the first push title set as the training target, training the first title generation model based on the second sample resource data until the first title generation model meets the model convergence condition to obtain the title generation model; the second sample resource data includes: the initial title and resource content of the first sample multimedia resources corresponding to the push titles in the first push title set.

[0009] Optionally, the title generation method further includes: training a second preset model based on the push titles of the second sample multimedia resources to obtain a title classification model; the push titles of the second sample multimedia resources include: positive sample push titles with a click-through rate greater than a preset threshold, and negative sample push titles with a click-through rate less than or equal to a preset threshold.

[0010] Optionally, the title generation method further includes: inputting the first push title into a pre-trained special effects generation model to obtain a second push title; the second push title includes the first push title and special effects resources.

[0011] Optionally, the title generation method further includes: obtaining the push title of a third sample multimedia resource; the push title of the third sample multimedia resource includes: a target push title and special effects resources; using the push title of the third sample multimedia resource as the training target, training a third preset model based on the target push title until the third preset model meets the model convergence condition, so as to obtain the special effects generation model.

[0012] According to a second aspect of the present disclosure, a title generation apparatus is provided, comprising: an acquisition unit and a first processing unit; the acquisition unit is configured to acquire target resource data of a target multimedia resource; the target resource data includes an initial title and resource content of the target multimedia resource; the first processing unit is configured to input the target resource data acquired by the acquisition unit into a pre-trained title generation model to obtain a first push title of the target multimedia resource; the title generation model is trained based on first sample resource data of a first sample multimedia resource; the first sample resource data includes: the initial title, push title, and resource content of the first sample multimedia resource.

[0013] Optionally, the title generation device further includes: a second processing unit; the second processing unit is configured to: input the push titles of the first sample multimedia resources into a pre-trained title classification model to obtain a first push title set and a second push title set; the click-through rate of the push titles in the first push title set is greater than the click-through rate of the push titles in the second push title set; perform a pairing operation on the first push title set and the second push title set according to a relevance algorithm to obtain multiple push title pairs; a push title pair includes: a push title in the first push title set, and a push title in the second push title set whose relevance to a push title in the first push title set is greater than a preset relevance; using the push titles in the first push title set among the multiple push title pairs as training targets, train a first preset model based on the push titles in the second push title set among the multiple push title pairs and the first resource content, until the first preset model meets the model convergence condition to obtain a first title generation model; the first resource content includes: the resource content of the first sample multimedia resources corresponding to the push titles in the second push title set among the multiple push title pairs; and determine the title generation model according to the first title generation model.

[0014] Optionally, the second processing unit is specifically used to: determine the first title generation model as the title generation model; or, using the push titles in the first push title set as the training target, train the first title generation model based on the second sample resource data until the first title generation model meets the model convergence condition to obtain the title generation model; the second sample resource data includes: the initial title and resource content of the first sample multimedia resources corresponding to the push titles in the first push title set.

[0015] Optionally, the second processing unit is further configured to: train the second preset model based on the push titles of the second sample multimedia resources to obtain a title classification model; the push titles of the second sample multimedia resources include: positive sample push titles with a click-through rate greater than a preset threshold, and negative sample push titles with a click-through rate less than or equal to a preset threshold.

[0016] Optionally, the first processing unit is further configured to: input the first push title into a pre-trained special effects generation model to obtain a second push title; the second push title includes the first push title and special effects resources.

[0017] Optionally, the second processing unit is further configured to: obtain the push title of the third sample multimedia resource; the push title of the third sample multimedia resource includes: target push title and special effects resource; using the push title of the third sample multimedia resource as the training target, train the third preset model based on the target push title until the third preset model meets the model convergence condition, so as to obtain the special effects generation model.

[0018] According to a third aspect of the present disclosure, a title generation apparatus is provided, which may include: a processor and a memory for storing processor-executable instructions; wherein the processor is configured to execute instructions to implement any of the optional title generation methods of the first aspect described above.

[0019] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, on which instructions are stored, such that when the instructions in the computer-readable storage medium are executed by a processor of a title generation apparatus, the title generation apparatus is able to perform any of the optional title generation methods of the first aspect described above.

[0020] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising instructions that, when executed on a processor of an electronic device, cause the electronic device to perform any of the optional title generation methods of the first aspect described above.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure.

[0022] The technical solution provided in this disclosure brings at least the following beneficial effects:

[0023] Based on any of the above aspects, in this disclosure, the resource push terminal can input the target resource data of the acquired target multimedia resource into a pre-trained title generation model to obtain the first push title of the target multimedia resource. The target resource data includes the initial title and resource content of the target multimedia resource. The title generation model is trained based on the initial title, push title, and resource content of the first sample multimedia resource. Compared to general technologies, the solution provided by this disclosure, by using a title generation model, is not affected by human factors (e.g., holidays, personnel changes, etc.), thus solving the technical problem of high cost and low efficiency in title generation when the workload is large in general technologies. Attached Figure Description

[0024] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0025] Figure 1 A schematic diagram of a title generation system provided in an embodiment of this disclosure;

[0026] Figure 2 This illustration shows a flowchart of a title generation method provided in an embodiment of the present disclosure. Figure 1 ;

[0027] Figure 3 This illustration shows a flowchart of a title generation method provided in an embodiment of the present disclosure. Figure 2 ;

[0028] Figure 4 This illustration shows a flowchart of a title generation method provided in an embodiment of the present disclosure. Figure 3 ;

[0029] Figure 5 This illustration shows a flowchart of a title generation method provided in an embodiment of the present disclosure. Figure 4 ;

[0030] Figure 6 This illustration shows a flowchart of a title generation method provided in an embodiment of the present disclosure. Figure 5 ;

[0031] Figure 7 This illustration shows a flowchart of a title generation method provided in an embodiment of the present disclosure. Figure 6 ;

[0032] Figure 8 This illustration shows a display diagram of a resource push terminal provided in an embodiment of the present disclosure;

[0033] Figure 9 This illustration shows a flowchart of a title generation method provided in an embodiment of the present disclosure. Figure 7 ;

[0034] Figure 10 This illustration shows a flowchart of a title generation method provided in an embodiment of the present disclosure. Figure 8 ;

[0035] Figure 11 A schematic diagram of the structure of a title generation apparatus provided in an embodiment of this disclosure is shown. Figure 1 ;

[0036] Figure 12 A schematic diagram of the structure of a title generation apparatus provided in an embodiment of this disclosure is shown. Figure 2 . Detailed Implementation

[0037] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0038] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0039] It should also be understood that the term "comprising" indicates the presence of the described feature, whole, step, operation, element and / or component, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements and / or components.

[0040] The data disclosed herein may be data authorized by the user or fully authorized by all parties.

[0041] As described in the background section, the common method is to generate push titles for multimedia resources through manual editing, combined with subjective judgment and understanding. When the workload is large, this method suffers from problems such as high cost and low efficiency.

[0042] Based on this, this disclosure provides a title generation method. The resource push terminal can input the target resource data of the acquired target multimedia resource into a pre-trained title generation model to obtain the first push title of the target multimedia resource. The target resource data includes the initial title and resource content of the target multimedia resource. The title generation model is trained based on the initial title, push title, and resource content of a first sample multimedia resource. Compared to general technologies, the solution provided by this disclosure, by using a title generation model, is not affected by human factors (e.g., holidays, personnel changes), thus solving the technical problem of high cost and low efficiency in title generation when the workload is large in general technologies.

[0043] The title generation method provided in this disclosure is illustrated below with reference to the accompanying drawings:

[0044] Figure 1 A schematic diagram of a title generation system provided in this disclosure embodiment is shown below. Figure 1As shown, the title generation system may include a resource push terminal 110 and a resource request terminal 120. The resource push terminal 110 and the resource request terminal 120 are connected.

[0045] Optionally, resource push client 110 can connect to multiple resource request clients, and resource request client 120 can also connect to multiple resource push clients. For ease of understanding, Figure 1 The following example illustrates the connection between a resource push client 110 and a resource request client 120.

[0046] Specifically, the resource push terminal 110 can determine the push title of the multimedia resource to be pushed and send the push title to the resource request terminal 120. In response to the click operation of the push title, the resource request terminal 120 can request the multimedia resource from the resource push terminal 110.

[0047] In some embodiments, the resource push terminal 110 can be a device for publishing multimedia resources, a chip within that device, or a system-on-a-chip within that device. The device can be any electronic product capable of human-computer interaction with a user through one or more methods such as a keyboard, touchpad, touchscreen, remote control, voice interaction, or handwriting device, such as a mobile phone, tablet computer, PDA, personal computer (PC), wearable device, smart TV, etc.

[0048] In some embodiments, the resource push terminal 110 can also be a device used by the application platform to manage multimedia resources, a chip in the device, or a system-on-a-chip in the device. The device can be a single server, or a server cluster consisting of multiple servers.

[0049] In some embodiments, the resource requester 120 is a device requesting multimedia resources, or it can be a chip within the device, or it can be a system-on-a-chip within the device. The device can be any electronic product capable of human-computer interaction with a user through one or more methods such as a keyboard, touchpad, touchscreen, remote control, voice interaction, or handwriting device, such as a mobile phone, tablet computer, PDA, PC, wearable device, smart TV, etc.

[0050] It should be noted that both the resource push terminal 110 and the resource request terminal 120 mentioned above can be referred to as electronic devices.

[0051] The title generation method provided in this disclosure can be applied to the aforementioned... Figure 1The resource push terminal 110 is shown in the application scenario. The title generation method provided in this embodiment includes: a process in which the resource push terminal 110 trains a title classification model based on the push title of a second sample multimedia resource to obtain a title classification model (hereinafter referred to as "title classification model training process"); a process in which the resource push terminal 110 trains a title generation model based on the sample resource data of a first sample multimedia resource (hereinafter referred to as "title generation model training process"); a process in which the resource push terminal 110 inputs target resource data into the title generation model to obtain the first push title of the target multimedia resource (hereinafter referred to as "title generation process"); and a process in which the resource push terminal 110 trains a special effects generation model based on the push title of a third sample multimedia resource (hereinafter referred to as "special effects generation model training process") and a process in which the resource push terminal 110 inputs the first push title into the special effects generation model to obtain the second push title (hereinafter referred to as "special effects generation process").

[0052] The following describes the "heading classification model training process," which includes the following heading generation method:

[0053] S1. The resource push terminal trains the second preset model based on the push title of the second sample multimedia resource to obtain the title classification model.

[0054] The push titles of the second sample multimedia resources include: positive sample push titles with a click-through rate greater than a preset threshold, and negative sample push titles with a click-through rate less than or equal to a preset threshold.

[0055] Optionally, the second sample multimedia resources may include multiple multimedia resources pushed within the same historical time period.

[0056] Optionally, multimedia resources may include: image resources, text resources, audio resources, video resources, etc.

[0057] Optionally, the push title of the second sample multimedia resource can be a push title generated by manual editing, or a push title randomly generated based on the resource content.

[0058] Optionally, the second preset model can be a model with self-learning function, such as a back propagation neural network model, a radial basis function (RBF) neural network model, a perceptron neural network model, a linear neural network model, a self-organizing neural network model, or a feedback neural network model.

[0059] For example, the preset push titles of the second sample multimedia resources include: Title A and Title B. Title A has a click-through rate of 50% in the past 24 hours, and Title B has a click-through rate of 10% in the past 24 hours. When the preset threshold is 30%, the resource pusher determines Title A as a positive sample push title and Title B as a negative sample push title.

[0060] Then, the resource push client uses title A and title B as sample data to train a second preset model to obtain a title classification model.

[0061] The technical solution provided by the above embodiments brings at least the following beneficial effects: As shown in S1, the resource push terminal can train the second preset model based on the push title of the second sample multimedia resource to obtain a title classification model. Since the push title of the second sample multimedia resource includes: positive sample push titles with a click-through rate greater than a preset threshold and negative sample push titles with a click-through rate less than or equal to the preset threshold, the trained title classification model can predict whether the input push title is a push title with a high click-through rate or a push title with a low click-through rate.

[0062] The following describes the "Title Generation Model Training Process," such as... Figure 2 As shown, after the resource push client obtains the title classification model, the title generation method can include:

[0063] S201. The resource push terminal inputs the push title of the first sample multimedia resource into the pre-trained title classification model to obtain the first push title set and the second push title set.

[0064] Among them, the click-through rate of push titles in the first set of push titles is greater than that of push titles in the second set of push titles.

[0065] Optionally, the first sample multimedia resources may include multiple multimedia resources pushed in the past.

[0066] Optionally, the push title of the first sample multimedia resource can be a push title generated by manual editing, or a push title randomly generated based on the resource content.

[0067] Since the title classification model is trained based on positive samples with click-through rates (CTRs) greater than a preset threshold and negative samples with CTRs less than or equal to the preset threshold, it can predict whether the input title will have a high or low CTR. Therefore, when the resource pusher inputs the titles of the first sample multimedia resources into the title classification model, it can obtain a first set of push titles consisting of titles with high CTRs and a second set of push titles consisting of titles with low CTRs.

[0068] S202. The resource push terminal performs a pairing operation on the first push title set and the second push title set according to the relevance algorithm to obtain multiple push title pairs.

[0069] A push title pair includes: a push title from a first set of push titles, and a push title from a second set of push titles whose relevance to a push title from the first set of push titles is greater than a preset relevance.

[0070] Optionally, the relevance algorithm may include the best match (BM) algorithm of the 25th iteration (BM25 algorithm).

[0071] Optionally, the preset relevance can be determined based on the maximum relevance.

[0072] Specifically, the resource push client can determine a push title from the first set of push titles based on a relevance algorithm, and then assign multiple relevance scores to each of the multiple push titles in the second set of push titles. The resource push client can then sort these multiple relevance scores from largest to smallest to obtain a relevance score sequence, and determine any value less than the first value in the sequence but greater than or equal to the second value as a preset relevance score.

[0073] At this point, the resource push client can identify a push title corresponding to a relevance score greater than a preset relevance score from among multiple relevance scores. This title is the one in the second set of push titles with the highest relevance score to a push title in the first set of push titles. Since a higher relevance score between two push titles indicates greater similarity, the resource push client can pair a push title from the first set of push titles with a push title from the second set of push titles whose relevance score to a push title in the first set is greater than the preset relevance score, thereby obtaining multiple push title pairs.

[0074] For example, the first set of push titles (set A) includes: title a1, title a2, title a3, and the second set of push titles (set B) includes: title b1, title b2, title b3, title b4. As shown in Table 1, based on the relevance algorithm, the resource push client can determine the relevance between title a1 and title b1 as 0.51, title a1 and title b2 as 0.80, title a1 and title b3 as 0.24, title a1 and title b4 as 0.65, a2 and title b1 as 0.17, title a2 and title b2 as 0.53, title a2 and title b3 as 0.42, title a2 and title b4 as 0.76, a3 and title b1 as 0.84, title a3 and title b2 as 0.62, title a3 and title b3 as 0.36, and title a3 and title b4 as 0.40.

[0075] The resource pusher can determine that the preset relevance to title a1 is 0.79 (less than 0.80 and greater than 0.65), the preset relevance to title a2 is 0.75 (less than 0.76 and greater than 0.53), and the preset relevance to title a3 is 0.83 (less than 0.84 and greater than 0.62).

[0076] Then, the resource pusher can combine two titles from sets A and B, which have a relevance greater than the preset relevance, into push title pairs, resulting in title pairs a1-b2, a2-b4, and a3-b1.

[0077] Table 1

[0078]

[0079] S203. The resource push terminal uses the push titles in the first push title set as the training target among multiple push title pairs, and the push titles in the second push title set and the first resource content to train the first preset model until the first preset model meets the model convergence condition, so as to obtain the first title generation model.

[0080] The first resource content includes the resource content of the first sample multimedia resource corresponding to the push title in the second push title set among multiple push title pairs.

[0081] Optionally, the resource content may include: optical character recognition (OCR) information of multimedia resources, automatic speech recognition (ASR) information of multimedia resources, etc.

[0082] Optionally, the first preset model can be a sequence-to-sequence (seq2seq) model.

[0083] Optionally, the model convergence condition may include at least one of the following: the model error is less than a preset error threshold, the model weight change is less than a preset weight threshold, or the model iteration count is greater than a preset maximum iteration count.

[0084] Specifically, the resource push client uses the push title with the highest click-through rate among multiple push title pairs as the training target. It trains a first preset model based on the push title with the lowest click-through rate and the resource content of the first sample multimedia resource corresponding to the push title with the lowest click-through rate. This allows the trained first title generation model to rewrite the input push title with the lowest click-through rate and the corresponding resource content into a push title with the highest click-through rate.

[0085] S204. The resource push terminal determines the first title generation model as the title generation model.

[0086] Since the first title generation model can rewrite the input title with a low click-through rate and the corresponding resource content into a push title with a high click-through rate, the resource push end can identify the first title generation model as the title generation model.

[0087] The technical solution provided by the above embodiments brings at least the following beneficial effects: As shown in S201-S204, the resource push terminal can input the push title of the first sample multimedia resource into a pre-trained title classification model to obtain a first push title set and a second push title set. Then, the resource push terminal can perform a pairing operation on the first push title set and the second push title set according to a relevance algorithm to obtain multiple push title pairs. Using the push titles in the first push title set among the multiple push title pairs as the training target, the first preset model is trained based on the push titles in the second push title set among the multiple push title pairs and the first resource content to obtain a first title generation model. Subsequently, the resource push terminal can determine the title generation model based on the first title generation model. In this way, the trained title generation model can rewrite the input title with a low click-through rate and the corresponding resource content into a push title with a high click-through rate.

[0088] In one optional embodiment, after the resource pusher obtains the first title generation model, the method for determining the title generation model is as follows: Figure 2 Based on the illustrated method embodiments, this embodiment provides a possible implementation. In conjunction with... Figure 2 ,like Figure 3 As shown, following S203, this title generation method further includes:

[0089] S301, The resource push terminal determines the first title generation model as the title generation model.

[0090] Since the first title generation model can rewrite the input title with a low click-through rate and the corresponding resource content into a push title with a high click-through rate, the resource push end can identify the first title generation model as the title generation model.

[0091] In another optional embodiment, after the resource pusher obtains the first title generation model, the method for determining the title generation model is as follows: Figure 2 Based on the illustrated method embodiments, this embodiment provides a possible implementation. In conjunction with... Figure 2 ,like Figure 4 As shown, following S203, this title generation method further includes:

[0092] S401. The resource push terminal uses the push titles in the first push title set as the training target, and trains the first title generation model based on the second sample resource data until the first title generation model meets the model convergence condition, so as to obtain the title generation model.

[0093] The second sample resource data includes the initial title and resource content of the first sample multimedia resource corresponding to the push title in the first push title set.

[0094] Optionally, the initial title can be a push title generated by manual editing, or a push title randomly generated based on the resource content.

[0095] Optionally, resource content may include: video content, audio content, or animated GIFs, etc.

[0096] Specifically, after obtaining the first title generation model, the resource push client can use the push titles in the first set of push titles with high click-through rates as training targets. Based on the initial titles and content of the first sample multimedia resources corresponding to the push titles in the first set of push titles, the first title generation model can be further trained until it meets the model convergence condition, thus obtaining the title generation model. During this process, the parameters of the first title generation model can be further adjusted to improve its accuracy.

[0097] The technical solution provided by the above embodiments brings at least the following beneficial effects: As can be seen from S301 or S401, the resource push terminal can determine the first title generation model as the title generation model. Alternatively, the resource push terminal can use the push titles in the first push title set as the training target and train the first title generation model based on the second sample resource data to obtain the title generation model. In this way, the parameters of the first title generation model can be further adjusted to improve the accuracy of the title generation model.

[0098] The following describes the "title generation process," such as... Figure 5 As shown, the title generation method may include:

[0099] S501, The resource push terminal obtains the target resource data of the target multimedia resource.

[0100] The target resource data includes the initial title and resource content of the target multimedia resource.

[0101] Optionally, the target multimedia resource may include: the multimedia resource to be pushed now, or the multimedia resource with a click rate lower than a preset threshold from the multimedia resources pushed in the past.

[0102] In one possible implementation, when the resource pusher is a device that publishes the target multimedia resource, the method for the resource pusher to obtain the target resource data of the target multimedia resource may include: in response to an input operation by a user holding the device, determining the initial title of the target multimedia resource, and identifying the resource content of the target multimedia resource.

[0103] In another possible implementation, when the resource pusher is a device used by the application platform to manage target multimedia resources, the method for the resource pusher to obtain the target resource data of the target multimedia resources may include: receiving a resource publishing message sent by the device publishing the target multimedia resources. The resource publishing message requests the resource pusher to push the target multimedia resources and includes: the initial title and resource content of the target multimedia resources. The resource pusher obtains the target resource data of the target multimedia resources through the resource publishing message.

[0104] S502, The resource push terminal inputs the target resource data into the pre-trained title generation model to obtain the first push title of the target multimedia resource.

[0105] The title generation model is trained based on the first sample resource data of the first sample multimedia resources.

[0106] The first sample resource data includes: the initial title, push title, and resource content of the first sample multimedia resource.

[0107] Specifically, the title generation model can rewrite the input title with a low click-through rate and the corresponding resource content into a push title with a high click-through rate. Therefore, the resource pusher can input the initial title and resource content of the target multimedia resource into the title generation model to obtain the first push title with a high click-through rate.

[0108] For example, referring to Table 2, the initial title of the multimedia resource 1 obtained by the resource push terminal is "Character A and Character B are such a pair of clowns!", and the resource content of multimedia resource 1 includes: "Anime". By inputting the title generation model, the first push title can be obtained as "[Anime] Character A and Character B are such a pair of clowns!".

[0109] The resource pusher will obtain the initial title of multimedia resource 2 as "These queens' speeches are so satisfying!". The content of multimedia resource 2 includes "entertainment" and "variety shows". By inputting the title generation model, the first push title can be "The queens' speeches of female celebrities, I love them so much~".

[0110] The resource pusher will obtain the initial title of multimedia resource 3 as "Easy to Make! Recreate the Popular Golden Cheese Sticks with Zero Difficulty". The content of multimedia resource 3 includes: "Dessert", "Spring Food Showcase", and "Cheese". By inputting the title generation model, the first push title can be obtained as "Teach You How to Make Golden Cheese Sticks, Easy to Recreate with Zero Difficulty".

[0111] Table 2

[0112]

[0113] The technical solution provided by the above embodiments brings at least the following beneficial effects: As shown in S501-S502, the resource push terminal can input the target resource data of the acquired target multimedia resource into a pre-trained title generation model to obtain the first push title of the target multimedia resource. The target resource data includes the initial title and resource content of the target multimedia resource. The title generation model is trained based on the initial title, push title, and resource content of the first sample multimedia resource. Compared with general technologies, the solution provided by this disclosure, by using a title generation model, is not affected by human factors (e.g., holidays, personnel changes, etc.), solving the technical problem of high cost and low efficiency in title generation when the workload is large in general technologies.

[0114] The following describes the "Special Effects Generation Model Training Process," such as... Figure 6 As shown, the title generation method may include:

[0115] S601, The resource push terminal obtains the push title of the third sample multimedia resource.

[0116] The push titles for the multimedia resources in the third sample include: the target push title and special effects resources.

[0117] Optional special effects resources may include: emoticons, filters, background music, animations, etc., such as emojis.

[0118] S602. The resource push terminal uses the push title of the third sample multimedia resource as the training target, and trains the third preset model based on the target push title until the third preset model meets the model convergence condition, so as to obtain the special effects generation model.

[0119] Optionally, the third preset model can be a seq2seq model.

[0120] When the third preset model is a seq2seq model, the seq2seq model can be trained using the pre-training mode (bart).

[0121] Specifically, the resource push client uses the push title of a third sample multimedia resource, which includes the target push title and special effects resources, as the training target. It trains a third preset model based on the target push title, and the resulting special effects generation model can add special effects resources to push titles that do not include special effects resources.

[0122] In one possible approach, the method by which the resource push client obtains the target push title may include: after obtaining the push title of the third sample multimedia resource, the resource push client can mask the special effects resources in the push title of the third sample multimedia resource and determine the unmasked push title as the target push title.

[0123] In another possible approach, the method for the resource push client to obtain the target push title may include: after obtaining the push title of the third sample multimedia resource, the resource push client can call a pre-created text extraction tool to identify the text portion from the push title of the third sample multimedia resource and determine the identified text portion as the target push title.

[0124] The technical solution provided by the above embodiments brings at least the following beneficial effects: As shown in S601-S602, the resource push terminal can determine the target push title excluding special effects resources based on the push title of the third sample multimedia resource. Then, the resource push terminal uses the push title of the third sample multimedia resource as the training target and trains the third preset model based on the target push title to obtain the special effects generation model. In this way, the special effects generation model obtained by the resource push terminal can add special effects resources to push titles that do not include special effects resources.

[0125] The following describes the "special effects generation process," combined with... Figure 5 ,like Figure 7 As shown, after the resource push terminal obtains the first push title of the target multimedia resource, the title generation method may further include:

[0126] S701. The resource push terminal inputs the first push title into the pre-trained special effects generation model to obtain the second push title.

[0127] The second push title includes the first push title and special effects resources.

[0128] Specifically, since the special effects generation model can add special effects resources to push titles that do not include special effects resources, the resource push terminal inputs the first push title that does not include special effects resources into the special effects generation model, and can obtain a second push title that includes the first push title and special effects resources.

[0129] For example, in combination Figure 8 The resource push client obtains the initial title 801 of the target multimedia resource as "Today, let's make a new way to eat eggplant," and the resource content 802 of the target multimedia resource. After inputting the initial title 801 and resource content 802 into the title generation model, the resource push client can obtain the first push title 803 as "A new way to eat eggplant, delicious and perfect with rice~." Then, the resource push client inputs the first push title 803 into the special effects generation model to obtain the second push title 804.

[0130] The technical solution provided by the above embodiments brings at least the following beneficial effects: As shown in S701, the resource push terminal inputs the first push title into the pre-trained special effects generation model to obtain a second push title that includes the first push title and special effects resources. This increases the appeal of the push title, making it easier to attract user interest and further increasing the click-through rate of the target multimedia resources.

[0131] The following is combined with Figure 9 and Figure 10 The embodiments of this disclosure will be described.

[0132] like Figure 9 As shown, the resource push terminal uses positive sample push titles with a click rate greater than a preset threshold and negative sample push titles with a click rate less than or equal to the preset threshold as sample data from the push titles of the second sample multimedia resources to train a title classification model.

[0133] Then, the resource push client inputs the push titles of the first sample multimedia resources into the title classification model to obtain a first set of push titles and a second set of push titles. The click-through rate (CTR) of the push titles in the first set is greater than that of the push titles in the second set.

[0134] The resource push client trains a title generation model that takes the push titles and corresponding resource content in the second push title set as input and the push titles in the first push title set as output.

[0135] like Figure 10 As shown, the resource push client trains a special effects generation model that takes push titles excluding special effects resources as input and push titles including special effects resources as output.

[0136] It is understood that, in practical implementation, the title generation apparatus described in the embodiments of this disclosure may include one or more hardware structures and / or software modules for implementing the aforementioned corresponding title generation method, and these hardware structures and / or software modules may constitute an electronic device. Those skilled in the art should readily recognize that, in conjunction with the algorithm steps of the various examples described in connection with the embodiments disclosed herein, this disclosure can be implemented in hardware or a combination of hardware and computer software. Whether a function is implemented in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.

[0137] Based on this understanding, this disclosure also provides a title generation device. Figure 11 A schematic diagram of the structure of the title generation apparatus provided in the embodiments of this disclosure is shown. Figure 1 .like Figure 11 As shown, the title generation device includes: an acquisition unit 1101 and a first processing unit 1102.

[0138] The acquisition unit 1101 is used to acquire target resource data of the target multimedia resource. For example, combined with... Figure 5 The acquisition unit 1101 is used to execute S501.

[0139] The first processing unit 1102 is used to input the target resource data acquired by the acquisition unit 1101 into a pre-trained title generation model to obtain the first push title of the target multimedia resource. For example, combined with Figure 5 The first processing unit 1102 is used to execute S502.

[0140] Optionally, the title generation apparatus further includes a second processing unit 1103.

[0141] The second processing unit 1103 is configured to: input the push titles of the first sample multimedia resources into a pre-trained title classification model to obtain a first push title set and a second push title set; perform a pairing operation on the first push title set and the second push title set according to a relevance algorithm to obtain multiple push title pairs; use the push titles in the first push title set among the multiple push title pairs as training targets, and train a first preset model based on the push titles in the second push title set among the multiple push title pairs and the content of the first resource, until the first preset model meets the model convergence condition to obtain a first title generation model; and determine a title generation model based on the first title generation model. For example, combining... Figure 2 The second processing unit 1103 is used to execute S201-S205.

[0142] Optionally, the second processing unit 1103 is specifically used to: determine the first title generation model as the title generation model; or, using the push titles in the first push title set as the training target, train the first title generation model based on the second sample resource data until the first title generation model meets the model convergence condition, so as to obtain the title generation model. For example, combined with Figure 3 The second processing unit 1103 is used to execute S301 or S401.

[0143] Optionally, the second processing unit 1103 is further configured to: train a second preset model based on the push titles of the second sample multimedia resources to obtain a title classification model. For example, combined with Figure 2 The second processing unit 1103 is used to execute S201.

[0144] Optionally, the first processing unit 1102 is further configured to: input the first push title into a pre-trained special effects generation model to obtain a second push title. For example, combined with Figure 7 The first processing unit 1102 is used to execute S701.

[0145] Optionally, the second processing unit 1103 is further configured to: obtain the push title of the third sample multimedia resource; use the push title of the third sample multimedia resource as the training target, and train the third preset model based on the target push title until the third preset model meets the model convergence condition, so as to obtain the special effects generation model. For example, combined with Figure 6 The second processing unit 1103 is used to execute S601-S602.

[0146] As described above, the present disclosure embodiments can divide the title generation device into functional modules according to the above method examples. The integrated modules can be implemented in hardware or as software functional modules. Furthermore, it should be noted that the module division in the present disclosure embodiments is illustrative and only represents one logical functional division; in actual implementation, other division methods may be used. For example, each function can be divided into its own functional modules, or two or more functions can be integrated into one processing module.

[0147] Regarding the title generation device in the above embodiments, the specific methods of operation of each module and its beneficial effects have been described in detail in the foregoing method embodiments, and will not be repeated here.

[0148] This disclosure also provides a title generation apparatus. Figure 12 A schematic diagram of the structure of the title generation apparatus provided in the embodiments of this disclosure is shown. Figure 2 The title generation apparatus may include at least one processor 221, a communication bus 222, a memory 223, and at least one communication interface 224.

[0149] Processor 221 may be a processor (central processing unit, CPU), microprocessor unit, ASIC, or one or more integrated circuits for controlling the execution of programs according to the present disclosure. Figure 11 The processor 221 is used to execute the operations performed by the acquisition unit 1101, the first processing unit 1102, and the second processing unit 1103.

[0150] The communication bus 222 may include a path for transmitting information between the aforementioned components.

[0151] Communication interface 224 uses any transceiver-like device for communicating with other devices or communication networks, such as electronic devices, Ethernet, radio access network (RAN), wireless local area network (WLAN), etc.

[0152] The memory 223 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory may exist independently and be connected to the processing unit via a bus. The memory may also be integrated with the processing unit.

[0153] The memory 223 stores the application code that executes the scheme of this disclosure, and its execution is controlled by the processor 221. The processor 221 executes the application code stored in the memory 223 to implement the functions of the method of this disclosure.

[0154] In a specific implementation, as one embodiment, processor 221 may include one or more CPUs, for example... Figure 12 CPU0 and CPU1 in the CPU.

[0155] In a specific implementation, as one example, the title generation device may include multiple processors, such as... Figure 12 Processors 221 and 225 are described herein. Each of these processors may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. A processor here may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0156] In a specific implementation, as one embodiment, the title generation device may further include an input device 226 and an output device 227. The input device 226 and output device 227 communicate and can accept user input in various ways. For example, the input device 226 may be a mouse, keyboard, touchscreen device, or sensing device. The output device 227 communicates with the processor 221 and can display information in various ways. For example, the output device 227 may be a liquid crystal display (LCD), a light emitting diode (LED) display device, etc.

[0157] Those skilled in the art will understand that Figure 12 The structure shown does not constitute a limitation on the title generation device, and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0158] This disclosure also provides a computer-readable storage medium including instructions stored thereon, which, when executed by a processor of an electronic device, enable the electronic device to perform the header generation method provided in the embodiments described above. For example, the computer-readable storage medium may be a memory 223 including instructions, which may be executed by a processor 221 of an electronic device to perform the method described above.

[0159] Optionally, the computer-readable storage medium may be a non-transitory computer-readable storage medium, such as ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device.

[0160] This disclosure also provides a computer program product containing instructions that, when run on an electronic device, cause the electronic device to perform the title generation method provided in the embodiments described above.

[0161] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0162] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A title generation method, characterized in that, include: Obtain the target resource data of the target multimedia resource; The target resource data includes the initial title and resource content of the target multimedia resource; The target resource data is input into a pre-trained title generation model to obtain the first push title of the target multimedia resource; The title generation model is trained based on the first sample resource data of the first sample multimedia resource. The first sample resource data includes: the initial title, push title, and resource content of the first sample multimedia resource; The training method for the title generation model includes: The push titles of the first sample multimedia resources are input into a pre-trained title classification model to obtain a first push title set and a second push title set; the click-through rate of the push titles in the first push title set is greater than the click-through rate of the push titles in the second push title set. The first set of push titles and the second set of push titles are paired according to a relevance algorithm to obtain multiple push title pairs; a push title pair includes: a push title in the first set of push titles, and a push title in the second set of push titles whose relevance to a push title in the first set of push titles is greater than a preset relevance. Using the push titles in the first set of push titles among the plurality of push title pairs as the training target, the first preset model is trained based on the push titles in the second set of push titles among the plurality of push title pairs and the first resource content, until the first preset model meets the model convergence condition, so as to obtain the first title generation model; the first resource content includes: the resource content of the first sample multimedia resource corresponding to the push titles in the second set of push titles among the plurality of push title pairs; The title generation model is determined based on the first title generation model.

2. The title generation method according to claim 1, characterized in that, The step of determining the title generation model based on the first title generation model includes: The first title generation model is determined as the title generation model; or... Using the push titles in the first push title set as the training target, the first title generation model is trained based on the second sample resource data until the first title generation model meets the model convergence condition, so as to obtain the title generation model; the second sample resource data includes: the initial title and resource content of the first sample multimedia resource corresponding to the push title in the first push title set.

3. The title generation method according to claim 1, characterized in that, Also includes: The second preset model is trained based on the push titles of the second sample multimedia resources to obtain the title classification model; The push titles of the second sample multimedia resources include: positive sample push titles with a click-through rate greater than a preset threshold, and negative sample push titles with a click-through rate less than or equal to the preset threshold.

4. The title generation method according to claim 1, characterized in that, Also includes: The first push title is input into a pre-trained special effects generation model to obtain a second push title; the second push title includes the first push title and special effects resources.

5. The title generation method according to claim 4, characterized in that, Also includes: Obtain the push title of the third-sample multimedia resource; The push title of the third sample multimedia resource includes: the target push title and the special effects resource; Using the push title of the third sample multimedia resource as the training target, the third preset model is trained based on the target push title until the third preset model meets the model convergence condition, so as to obtain the special effects generation model.

6. A title generation device, characterized in that, include: Acquisition unit and first processing unit; The acquisition unit is used to acquire target resource data of the target multimedia resource; The target resource data includes the initial title and resource content of the target multimedia resource; The first processing unit is used to input the target resource data acquired by the acquisition unit into a pre-trained title generation model to obtain the first push title of the target multimedia resource; The title generation model is trained based on the first sample resource data of the first sample multimedia resource. The first sample resource data includes: the initial title, push title, and resource content of the first sample multimedia resource; The training method for the title generation model includes: The push titles of the first sample multimedia resources are input into a pre-trained title classification model to obtain a first push title set and a second push title set; the click-through rate of the push titles in the first push title set is greater than the click-through rate of the push titles in the second push title set. The first set of push titles and the second set of push titles are paired according to a relevance algorithm to obtain multiple push title pairs; a push title pair includes: a push title in the first set of push titles, and a push title in the second set of push titles whose relevance to a push title in the first set of push titles is greater than a preset relevance. Using the push titles in the first set of push titles among the plurality of push title pairs as the training target, the first preset model is trained based on the push titles in the second set of push titles among the plurality of push title pairs and the first resource content, until the first preset model meets the model convergence condition, so as to obtain the first title generation model; the first resource content includes: the resource content of the first sample multimedia resource corresponding to the push titles in the second set of push titles among the plurality of push title pairs; The title generation model is determined based on the first title generation model.

7. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the title generation method as described in any one of claims 1-5.

8. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is enabled to perform the title generation method as described in any one of claims 1-5.

9. A computer program product, comprising instructions, characterized in that, When the instructions are executed on the processor of the electronic device, the electronic device causes the electronic device to perform the title generation method as described in any one of claims 1-5.

Images