CN116033207A

Movatterモバイル変換

Info

Publication number: CN116033207A
Application number: CN202211584288.2A
Authority: CN
Inventors: 于洋
Original assignee: Beijing QIYI Century Science and Technology Co Ltd
Current assignee: Beijing QIYI Century Science and Technology Co Ltd
Priority date: 2022-12-09
Filing date: 2022-12-09
Publication date: 2023-04-28
Anticipated expiration: 2042-12-09
Also published as: CN116033207B

Abstract

The embodiment of the invention provides a method and a device for generating a video title, electronic equipment and a readable storage medium, wherein the method comprises the following steps: acquiring at least one barrage of a target video; scoring the at least one bullet screen according to target scoring parameters, the target scoring parameters comprising: bullet screen quality, interaction data and content correlation; and under the condition that the score of the target barrage in the at least one barrage reaches the expected score, inputting the target barrage into a pre-trained deep learning model to obtain a video title of the target video. According to the video title generation method, the barrages in the target video are scored by means of the target scoring parameters, so that the target barrages with higher quality are screened out, and the target titles are determined through the target barrages due to higher correlation between barrage contents and video contents, so that the correlation between the video titles and the video contents is improved.

Description

Video title generation method and device, electronic equipment and readable storage medium

Technical Field

The present invention relates to the field of video processing technologies, and in particular, to a method and apparatus for generating a video title, an electronic device, and a readable storage medium.

Background

As short videos get hotter in the network, the number of short videos generated in various video platforms per day is also getting more and more, and thus, the workload of header creation for short videos is also getting more and more heavy. At present, besides manually labeling the title of the short video, the title can be automatically generated for the short video through a training model, but at present, the title of the short video can not be closely related to video content and user feedback through the automatic generation of the short video through the training model, so that the problem of low relevance between the video title and the video content occurs.

Disclosure of Invention

The embodiment of the invention aims to provide a method for generating video titles, which solves the problem that the video titles automatically generated in the prior art have lower attraction to users.

In a first aspect of the present invention, a training method for a model is provided: comprising the following steps:

acquiring a created language learning model, wherein the language learning model is used for inputting text content;

training the language learning model through a first training sample to obtain an output value, and updating the language learning model according to the output value, wherein the first training sample comprises a barrage sample and a speech sample;

And determining the trained language learning model as the deep learning model.

In a second aspect of the present invention, there is provided a method for generating a title, including:

acquiring at least one barrage of a target video;

scoring the at least one bullet screen according to target scoring parameters, the target scoring parameters comprising: bullet screen quality, interaction data and content correlation;

and under the condition that the score of the target barrage in the at least one barrage reaches the expected score, inputting the target barrage into a pre-trained deep learning model to obtain a video title of the target video.

Optionally, when the score of the target barrage in the at least one barrage reaches the expected score, inputting the target barrage into a pre-trained deep learning model to obtain a video title of the target video, including:

under the condition that the score of the target barrage in the at least one barrage reaches the expected score, text integration is carried out on the target barrage and the line text of the target video, and a target text is generated;

and inputting the target text into a pre-trained deep learning model to obtain a video title of the target video.

Optionally, the scoring the at least one barrage according to the target scoring parameter includes:

scoring the at least one bullet screen according to preset bullet screen quality, wherein the bullet screen quality is used for indicating the smoothness degree and the word number of the bullet screen;

under the condition that a first bullet screen set accords with the preset bullet screen quality, scoring the first bullet screen set according to preset interaction data, wherein the interaction data are used for indicating the praise of the bullet screen, and the first bullet screen set is a set formed by one or more bullet screens in the at least one bullet screen;

and under the condition that the second barrage set accords with the preset interaction data, scoring the second barrage set according to the preset content relativity, wherein the content relativity is used for indicating the relativity of the barrages and the lines in the target video, and the second barrage set is a set formed by one or more barrages in the first barrage set.

Optionally, the scoring the second barrage set according to the preset content relevance includes:

clustering the second bullet screen collection based on the station word correlation degree and the station word concentration degree to obtain a clustering result, wherein the clustering result is used for representing the station word correlation degree between each bullet screen in the second bullet screen collection and a target video;

And scoring each barrage in the second barrage set according to the clustering result.

Optionally, in the case that the score of the target barrage in the at least one barrage reaches the expected score, inputting the target barrage into a pre-trained deep learning model, and after obtaining the video title of the target video, the method further includes:

acquiring a real-time heat barrage, wherein the heat barrage is the barrage with the highest occurrence frequency in at least one barrage;

and inputting the video titles of the heat barrage and the target video into the pre-trained deep learning model to obtain corrected video titles.

In a third aspect of the present invention, there is also provided a title generating apparatus, including:

the model creation module is used for obtaining a created language learning model, and the language learning model is used for identifying the content of the input text;

the model training module is used for training the language learning model through a first training sample to obtain an output value, and updating the language learning model according to the output value, wherein the first training sample comprises a barrage sample and a speech sample;

and the model determining module is used for determining the trained language learning model as the deep learning model.

In a fourth aspect of the present invention, there is also provided a title generating apparatus, including:

the acquisition module is used for acquiring at least one bullet screen of the target video;

the scoring module is configured to score the at least one bullet screen according to target scoring parameters, where the target scoring parameters include: bullet screen quality, interaction data and content correlation;

the input module is used for inputting the target barrage into the pre-trained deep learning model to obtain a video title of the target video under the condition that the score of the target barrage in the at least one barrage reaches the expected score.

Optionally, the input module further includes:

the combining sub-module is used for combining the target barrage with the line text of the target video to generate a target text under the condition that the score of the target barrage in the at least one barrage reaches the expected score;

and the input sub-module is used for inputting the target text into a pre-trained deep learning model to obtain a video title of the target video.

In a third aspect of the embodiments of the present invention, there is also provided an electronic device, including a processor, a memory, and a program or instructions stored on the memory and executable on the processor, the program or instructions implementing the steps of the method for generating a video title according to any one of the first aspects when executed by the processor.

In a fourth aspect of embodiments of the present invention, there is also provided a computer readable storage medium having stored thereon a program or instructions which, when executed by a processor, implement the steps of the method for generating a video title as in any of the first aspects.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below.

Fig. 1 is a flowchart of a method for generating a video title according to an embodiment of the present invention;

FIG. 2 is a flow chart of a training method of a model according to an embodiment of the present invention;

fig. 3 is a schematic structural diagram of a video title generating apparatus according to an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a training device for a model according to an embodiment of the present invention;

fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, are intended to be within the scope of the present application.

Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although a flowchart depicts steps as a sequential process, many of the steps may be implemented in parallel, concurrently, or with other steps. Furthermore, the order of the steps may be rearranged. The process may be terminated when its operations are completed, but may have additional steps not included in the figures. The processes may correspond to methods, functions, procedures, subroutines, and the like.

Furthermore, the terms "first," "second," and the like, may be used herein to describe various directions, acts, steps, or elements, etc., but these directions, acts, steps, or elements are not limited by these terms. These terms are only used to distinguish one direction, action, step or element from another direction, action, step or element. For example, the first speed difference may be referred to as a second speed difference, and similarly, the second speed difference may be referred to as the first speed difference, without departing from the scope of the present application. Both the first speed difference and the second speed difference are speed differences, but they are not the same speed difference. The terms "first," "second," and the like, are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present invention, the meaning of "plurality" means at least two, for example, two, three, etc., unless specifically defined otherwise.

The embodiment of the application provides a method for generating a video title, as shown in fig. 1, the method comprises the following steps:

Step 101, at least one barrage of a target video is acquired.

In the present embodiment, the target video may be a long video, a short video, a clip, or the like, and the short video is exemplified in the present embodiment. The bullet screen is characterized in that at least one bullet screen exists in the target video, specifically, the bullet screen refers to comment subtitles popped up when the video is watched on a network, the bullet screen content of the bullet screen is generally related to the target video, and the bullet screen can be 'too wonderful', 'good and beautiful' and the like.

Step 102, scoring the at least one bullet screen according to target scoring parameters, wherein the target scoring parameters comprise: bullet screen quality, interaction data, and content relevance.

In this embodiment, the obtained barrage is scored by means of target scoring parameters, where the target scoring parameters include: bullet screen quality, interaction data, and content relevance. The bullet screen quality indicates the smoothness of the bullet screen, whether the bullet screen conforms to grammar, and the like. The interactive data refers to the praise or the number of occurrences of the bullet screen, for example, the bullet screen is praise by a plurality of users or the same bullet screen is sent by a plurality of users. The content relevance refers to the relevance of the barrage to the video content, for example, the relevance of the barrage to the video line or the video character. And the target barrage meeting the requirements can be accurately screened out through the target scoring parameters.

And step 103, under the condition that the score of the target barrage in the at least one barrage reaches the expected score, inputting the target barrage into a pre-trained deep learning model to obtain a video title of the target video.

In this embodiment, the deep learning model is a pre-trained deep learning model, and the deep learning model may be a language model or the like. The pre-trained deep learning model can input a video title matched with a target video after inputting the target bullet screen. The obtained target barrage is input into the pre-trained deep learning model, and the video title of the target video is related to the barrage, so that the barrage is the understanding of the user on the video content, has stronger style and commentary, and better improves the relevance between the video title and the video content.

The embodiment of the invention provides a method for generating a video title, which comprises the following steps: acquiring at least one barrage of a target video; scoring the at least one bullet screen according to target scoring parameters, the target scoring parameters comprising: bullet screen quality, interaction data and content correlation; and under the condition that the score of the target barrage in the at least one barrage reaches the expected score, inputting the target barrage into a pre-trained deep learning model to obtain a video title of the target video. According to the video title generation method, the barrages in the target video are scored by means of the target scoring parameters, so that the target barrages with higher quality are screened out, and the target titles are determined through the target barrages due to higher correlation between barrage contents and video contents, so that the correlation between the video titles and the video contents is improved.

combining the target barrage with the line text of the target video to generate a target text under the condition that the score of the target barrage in the at least one barrage reaches the expected score;

In this embodiment, the speech text is a speech set in the target video, that is, a set in which speech of all characters in the target video are integrated. Note that, in a short video, a bystander, a comment, or the like is also included in the speech set.

Specifically, all target barrages are scored through three aspects of barrage quality, interaction data and content relatedness, and the expected score is set to 90, and the target barrages are selected to be combined with the line text when the target barrages exceed or are equal to 90 to generate target texts. It should be noted that, the speech text and the target barrage in the target text are input at the same time, and the content is not fused or modified, but only the two texts are combined. Because the station caption and the station caption have a part of information describing the current scenario, the correlation between the attraction of the title and the video can be improved by combining the station caption and the station caption to generate the title.

In this embodiment, the bullet screen is scored sequentially by the preset bullet screen quality, the preset interaction data and the preset content relativity, and the subsequent scoring can be performed only if the previous scoring standard is met.

Firstly, scoring at least one bullet screen through preset bullet screen quality, wherein the bullet screen quality is used for indicating the smoothness degree and the word number of the bullet screen, specifically, judging the content contained in the bullet screen by depending on the bullet screen quality, and the bullet screen quality reflects whether the bullet screen accords with grammar, is too short, is smooth or not and the like. After scoring, classifying the barrages meeting the preset barrage quality into a first barrage set, and removing the rest barrages. For example, the shorter and smoother the barrage, the better the score of "6-pole" is than the score of "better" and it should be noted that, if the content of the barrage includes the nearest internet stem, the higher the score is, because the internet stem can be liked by many young people, and the interest of the title is improved.

Secondly, grading the barrages in the first barrage set through preset interaction data, wherein the content contained in the barrages is judged by the interaction data, the interaction data are used for indicating the praise of the barrages, specifically, the interaction data indicate the heat and praise of the barrages, for example, the number of times of occurrence of a barrage is large and the barrages are praise in a large amount, namely, the barrages are high in praise, and users are easy to attract. After scoring, classifying the barrages meeting the preset interaction data into a second barrage set, and removing the rest barrages. For example, when the barrage contains interesting terms or new stems of the internet, the barrage can receive more praise from users, so that the score of the barrage is higher, for example, the barrage such as 'you are stronger than the national feet' can be praise from most people, and the interactive data is higher.

And finally scoring the barrages in the second barrage set through the preset content correlation, wherein the content contained in the barrages is judged by the content correlation, the content correlation indicates the correlation between the barrages and the lines in the target video, specifically, the coincidence degree between the lines reflects the content correlation, when the correlation is higher, the barrages are more attached to the video content, and the generated title can be more accordant with the requirements.

In this embodiment, the line correlation degree reflects the fitting degree of the line and the barrage, and the line concentration degree reflects the occurrence times of the line, so that the barrage in the second barrage concentration can be better scored through the line correlation degree and the line concentration degree, and the scoring quality is ensured. The score of any two barrages in the second barrage set can be calculated by taking the relevancy of the barrages and the concentration of the barrages as scoring criteria, the two barrages can be used as the same cluster under the condition that the score reaches the preset score, and the two barrages cannot be clustered under the condition that the score does not reach the preset score. After scoring all any two of the barrages, a clustering result is obtained, which may include one or more clusters.

After clustering, a clustering result is obtained, and the clustering result is used for indicating the corresponding station caption relevance and station caption concentration of each bullet screen, so that each bullet screen in the second bullet screen collection is scored. It should be noted that, when more barrages are included in the clustering result, the correlation degree between barrages is higher, so that when the barrages are scored, the clustering can be scored, specifically, the score of one cluster reflects the scores of all barrages in one cluster, so that the scoring time is saved, and the scoring efficiency is improved.

In this embodiment, after the video title is generated, the video title may be updated according to the real-time barrage sent by the user, so that the video title is more attractive.

Specifically, in the video playing process, when a bullet screen with high praise or a large number of occurrences appears in the video, the bullet screen is determined as a hot bullet screen. And then inputting the original titles of the hot barrage and the target video into a trained deep learning model at the same time, and outputting a corrected video title which is more fit with the current viewing environment. In this embodiment, due to the rapid development of the network society, many new stems are endlessly formed, so that the video can be attractive by updating the video title through the real-time barrage, and the click rate of the user is higher.

In another embodiment, as shown in fig. 2, there is also provided a model training method, including:

step 201, acquiring a created language learning model, wherein the language learning model is used for identifying input text content;

step 202, training the language learning model through a first training sample to obtain an output value, and updating the language learning model according to the output value, wherein the first training sample comprises a barrage sample and a speech sample;

and 203, determining the trained language learning model as the deep learning model.

In this embodiment, the deep learning model is a trained language model, and the language model is a language abstract mathematical modeling performed according to language objective facts, which is a corresponding relationship. The first training samples comprise a large number of speech bullet screen samples and speech bullet screen samples, the created first language model is trained by using the training samples to obtain output values, parameters in the first language model are continuously adjusted through the output values, so that the speech texts can be better identified, the first language model can learn various speech bullet screens and the like conforming to conditions, and titles generated by the models are more attached to the speech bullet screens.

After training by a large number of first training samples, a trained language learning model is obtained, and the model can generate titles related to the barrage after being input into the barrage.

According to the video title generation method, the bullet curtains in the target video are scored by means of the target scoring parameters, so that the target bullet curtains with higher quality are screened out, the target titles are determined through the target bullet curtains, and the relevance between the video titles and the video content is improved.

The embodiment of the invention also provides adevice 300 for generating the video title, as shown in fig. 3, thedevice 300 for generating the title comprises:

anacquisition module 310, configured to acquire at least one bullet screen of a target video;

ascoring module 320, configured to score the at least one bullet screen according to target scoring parameters, where the target scoring parameters include: bullet screen quality, interaction data and content correlation;

and theinput module 330 is configured to input the target barrage into a pre-trained deep learning model to obtain a video title of the target video when the score of the target barrage in the at least one barrage reaches the expected score.

Optionally, theinput module 330 further includes:

Optionally, thescoring module 320 further includes:

the first evaluation sub-module is used for scoring the at least one bullet screen according to preset bullet screen quality, and the bullet screen quality is used for indicating the smoothness degree and the word number of the bullet screen;

the second evaluation sub-module is used for scoring the first bullet screen set according to preset interaction data under the condition that the first bullet screen set accords with the preset bullet screen quality, wherein the interaction data is used for indicating the praise number of the bullet screens, and the first bullet screen set is a set formed by one or more bullet screens in the at least one bullet screen;

and the third evaluation sub-module is used for scoring the second barrage set according to the preset content relevance under the condition that the second barrage set accords with the preset interaction data, wherein the content relevance is used for indicating the relevance of the barrages and the lines in the target video, and the second barrage set is a set formed by one or more barrages in the first barrage set.

Optionally, the third evaluation submodule further includes:

the clustering unit is used for clustering the second bullet screen collection based on the station word correlation degree and the station word concentration degree to obtain a clustering result, and the clustering result is used for representing the station word correlation degree between each bullet screen in the second bullet screen collection and the target video;

and the clustering scoring unit is used for scoring each barrage in the second barrage aggregate according to the clustering result.

Optionally, the method further comprises:

the bullet screen acquisition module is used for acquiring a real-time heat bullet screen, wherein the heat bullet screen is the bullet screen with the highest occurrence frequency in the at least one bullet screen;

According to the method and the device, the barrage in the target video is scored by means of the target scoring parameters, so that the target barrage with higher quality is screened out, the target title is determined through the target barrage, and the relevance between the video title and the video content is improved due to the fact that the barrage content and the video content are higher in relevance.

The embodiment of the invention also provides atraining device 400 of the model, as shown in fig. 4, thetraining device 400 of the model includes:

Amodel creation module 410, configured to obtain a created language learning model, where the language learning model is used to identify input text content;

themodel training module 420 is configured to train the language learning model through a first training sample, obtain an output value, and update the language learning model according to the output value, where the first training sample includes a barrage sample and a speech sample;

themodel determining module 430 is configured to determine a trained language learning model as the deep learning model.

According to the method and the device, the barrage in the target video is scored by means of the target scoring parameters, so that the target barrage with higher quality is screened out, and the target title is determined through the target barrage due to higher correlation between the barrage content and the video content, so that the correlation between the video title and the video content is improved.

Fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, where, as shown in fig. 5, theelectronic device 500 includes amemory 510 and aprocessor 520, and the number of theprocessors 520 in theelectronic device 500 may be one or more, and oneprocessor 520 is taken as an example in fig. 5; thememory 510,processor 520 in the server may be connected by a bus or other means, for example in fig. 5.

Thememory 510 is a computer readable storage medium, and may be used to store a software program, a computer executable program, and a module, such as program instructions/modules corresponding to the method for generating a title in the embodiment of the present invention, and theprocessor 520 executes the software program, instructions, and modules stored in thememory 510, thereby performing various functional applications and data processing of the server/terminal/server, that is, implementing the method for generating a video title as described above.

Wherein theprocessor 520 is configured to execute a computer program stored in thememory 510, and implement the following steps:

acquiring at least one barrage of a target video;

Or the following steps are realized:

acquiring a created language learning model, wherein the language learning model is used for identifying the content of an input text;

And determining the trained language learning model as the deep learning model.

In one embodiment, the computer program of the electronic device provided by the embodiment of the present invention is not limited to the above method operations, but may also perform related operations in the video title generation method or the model training method provided by any embodiment of the present invention.

Thememory 510 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, at least one application program required for functions; the storage data area may store data created according to the use of the terminal, etc. In addition,memory 510 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some examples,memory 510 may further include memory remotely located relative toprocessor 520, which may be connected to a server/terminal/server via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The embodiment of the invention also provides a storage medium containing computer executable instructions, which when executed by a computer processor, are used for executing a method for generating a video title, the method comprising:

acquiring at least one barrage of a target video;

Or performing a training method of a model, comprising:

and determining the trained language learning model as the deep learning model.

Of course, the storage medium containing the computer executable instructions provided in the embodiments of the present invention is not limited to the method operations described above, and may also perform the related operations in the method for generating a title provided in any embodiment of the present invention.

The computer-readable storage media of embodiments of the present invention may take the form of any combination of one or more computer-readable media. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

The computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, either in baseband or as part of a carrier wave. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

Program code embodied on a storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Computer program code for carrying out operations of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, smalltalk, C ++ and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or terminal. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computer (for example, through the Internet using an Internet service provider).

Note that the above is only a preferred embodiment of the present invention and the technical principle applied. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, while the invention has been described in connection with the above embodiments, the invention is not limited to the embodiments, but may be embodied in many other equivalent forms without departing from the spirit or scope of the invention, which is set forth in the following claims.

Claims

1. A method of training a model, comprising:

and determining the trained language learning model as the deep learning model.

2. A method for generating a video title, comprising:

Acquiring at least one barrage of a target video;

3. The method according to claim 2, wherein, in the case that the score of the target barrage in the at least one barrage reaches the expected score, inputting the target barrage into a pre-trained deep learning model to obtain a video title of the target video, including:

4. The method of claim 2, wherein scoring the at least one barrage according to a target scoring parameter comprises:

5. The method of claim 4, wherein scoring the second set of barrages according to a predetermined content relevance comprises:

6. The method according to claim 1, wherein in the case that the score of the target barrage in the at least one barrage reaches the expected score, the target barrage is input into a pre-trained deep learning model, and after obtaining the video title of the target video, the method further comprises:

7. A training device for a model, comprising:

8. A video title generation apparatus, comprising:

9. An electronic device comprising a processor, a memory and a program or instruction stored on the memory and executable on the processor, which when executed by the processor, performs the steps of the method of generating a video title as claimed in any one of claims 2 to 7, or the steps of the method of training a model as claimed in claim 1.

10. A computer-readable storage medium, characterized in that the readable storage medium has stored thereon a program or instructions which, when executed by a processor, implement the steps of the method of generating a video title as claimed in any one of claims 2 to 7, or the steps of the method of training a model as claimed in claim 1.