Video processing method and device based on artificial intelligence, equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence and financial science and technology, and discloses a video processing method and device based on artificial intelligence, equipment and a storage medium. The method comprises the steps of obtaining an insurance consultation text language input by a user, wherein the insurance consultation text language is input by the user for an insurance product marketing video; inputting the insurance consultation text language and the insurance product marketing video into a pre-trained time activity positioning model, and cutting out a marketing video clip corresponding to the insurance consultation text language from the insurance product marketing video; the time activity positioning model is obtained by training based on training sample data, the training sample data comprises multiple sample insurance product marketing videos, each sample insurance product marketing video comprises a marketing video clip and sentence description of the marketing video clip, and the sentence description is used for representing starting and ending statements of the marketing video clip; the marketing video clips of the multiple sample insurance product marketing videos correspond to different insurance consultation types respectively.

Description

Technical Field

[0001] The present application relates to the fields of financial technology and data processing technology, and in particular to an artificial intelligence-based video processing method, device, equipment and storage medium. Background Art

[0002] Nowadays, there are many types of insurance products, including insurance products for children, middle-aged people, and the elderly. The dimensions of type classification are also different, and the types are huge. In order to introduce insurance products well, they are generally promoted and marketed through various channels and forms, among which the marketing form includes insurance product marketing videos.

[0003] The inventors have discovered that due to the variety of insurance product types, the corresponding insurance product marketing video content formats are also diverse. As insurance products involve a large amount of information, users generally manually drag and drop video searches to locate the desired video clips. It is difficult for users to quickly locate the video content portion they want from an insurance product marketing video, and the efficiency of locating the desired video clips of insurance product marketing videos is low. Summary of the invention

[0004] The embodiments of the present application provide a video processing method, apparatus, device and storage medium based on artificial intelligence to quickly locate desired video clips of insurance product marketing videos.

[0005] In a first aspect, a video processing method based on artificial intelligence is provided, comprising: Acquire the insurance consultation text language input by the user, where the insurance consultation text language is input by the user for the insurance product marketing video; Inputting the insurance consulting text language and the insurance product marketing video into a pre-trained time activity positioning model to cut out marketing video clips corresponding to the insurance consulting text language from the insurance product marketing video; Among them, the trained time-activity positioning model is trained based on training sample data, and the training sample data includes multiple sample insurance product marketing videos, each of the sample insurance product marketing videos includes a marketing video clip and a sentence description of the marketing video clip, and the sentence description is used to characterize the start and end sentences of the marketing video clip, and the marketing video clips of the multiple sample insurance product marketing videos correspond to different insurance consultation types.

[0006] In a second aspect, a video processing device based on artificial intelligence is provided, comprising: An acquisition module, used to acquire the insurance consultation text language input by the user, wherein the insurance consultation text language is input by the user for the insurance product marketing video; An input module, used to input the insurance consulting text language and the insurance product marketing video into a pre-trained time activity positioning model, so as to cut out the marketing video clip corresponding to the insurance consulting text language from the insurance product marketing video; Among them, the trained time-activity positioning model is trained based on training sample data, and the training sample data includes multiple sample insurance product marketing videos, each of the sample insurance product marketing videos includes a marketing video clip and a sentence description of the marketing video clip, and the sentence description is used to characterize the start and end sentences of the marketing video clip, and the marketing video clips of the multiple sample insurance product marketing videos correspond to different insurance consultation types. In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the artificial intelligence-based video processing method as described in any one of the preceding items is implemented.

[0007] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the video processing method based on artificial intelligence as described above is performed.

[0008] In one of the solutions provided in the present application, the insurance consulting text language input by the user is first obtained, and the insurance consulting text language is input by the user for the insurance product marketing video. Then, the insurance consulting text language and the insurance product marketing video are input into a pre-trained time-activity positioning model to cut out the marketing video clip corresponding to the insurance consulting text language from the insurance product marketing video. In this way, the user can quickly locate the video content part he wants from the insurance product marketing video, thereby improving the low efficiency of locating the desired video clip of the insurance product marketing video, thereby improving the user experience and the marketing conversion rate of the insurance product. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0010] Figure 1 This is a schematic diagram of an application environment of a video processing method based on artificial intelligence in one embodiment of the present application; Figure 2It is a schematic diagram of a network structure of a temporal activity positioning model in a video processing method based on artificial intelligence in one embodiment of the present application; Figure 3 This is a schematic diagram of a training process of a time activity positioning model in an embodiment of the present application; Figure 4 is another flow chart of a video processing method based on artificial intelligence in one embodiment of the present application; Figure 5 is another flow chart of a video processing method based on artificial intelligence in one embodiment of the present application; Figure 6 is a structural schematic diagram of a video processing device based on artificial intelligence in one embodiment of the present application; Figure 7 It is a structural diagram of a computer device in one embodiment of the present application. DETAILED DESCRIPTION

[0011] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0012] The embodiment of the present application provides a video processing method based on artificial intelligence. The video processing method based on artificial intelligence can be applied as follows: Figure 1 Specifically, the application environment includes: Figure 1 The client shown can be installed on, but not limited to, various personal computers, laptops, smart phones, tablet computers, and portable wearable devices. The client is used to implement a video processing method based on artificial intelligence.

[0013] In this embodiment, a temporal activity localization model (Temporal Activity Localization via Language, TALL) needs to be trained first. The trained temporal activity localization model is trained based on training sample data. The training sample data includes multiple sample insurance product marketing videos. Each of the sample insurance product marketing videos includes a marketing video clip and a sentence description of the marketing video clip. The sentence description is used to characterize the start and end sentences of the marketing video clip. The marketing video clips of the multiple sample insurance product marketing videos correspond to different insurance consultation types. To facilitate understanding of the embodiments of the present application, the temporal activity localization model provided in the embodiments of the present application and its training process are first described.

[0014] like Figure 2 As shown, Figure 2 A network framework diagram and a processing diagram of a temporal activity localization model provided in an embodiment of the present application are provided to facilitate understanding of the data processing process of the subsequent model training and application. The temporal activity localization model mainly includes a visual encoder, a sentence encoder, a multi-modal processing module, a temporal localization regression network, and a multi-task loss function, among which: Visual Encoder For a long untrimmed video , a set of video clips is generated through a time sliding window ,in Yes Video Total number of clips, and It's a video clip The start and end time of the visual encoder in this model is defined as a function , used to convert a video clip and its context is mapped to a feature vector , whose dimension is Inside the visual encoder, the feature extractor Extract clip-level feature vector, whose input is Frame, output is dimension Vector. Video clip , , is the displacement of the clip, is the displacement boundary, that is, the video clip It takes itself as the center clip and the surrounding clips as context clips.

[0015] In this model, uniform sampling is performed from each video segment (center segment and context segment) The feature vector of the central segment is recorded as For the context fragment, a pooling layer is used to calculate the previous context features and post-context The pre-context features and post-context features are separated because the end and start of an activity in a video clip may be very different and both are important for temporal localization. , and After connecting them, they are linearly transformed into a dimension of The eigenvector of , as a video clip Visual representation of feature vector.

[0016] Sentence Encoder A sentence encoder is a function , used to describe the video clip in sentences Mapped to dimension (same as visual feature space). Specifically, we use the sentence embedding extractor Extracting sentence-level embeddings , and then through the linear transformation layer Mapped to dimension of , and the visual representation feature vector same.

[0017] In the embodiment of the present application, the model combines two sentence embedding extractors, one is a recurrent neural network (Long Short-Term Memory, LSTM), which takes a word in the input text language as input in each step, and the hidden state of the last step is used as the sentence-level Embedding, and the other is a sentence encoder BERT (for example, the model training is performed using skip-thoughts, but the model structure is consistent with the general BERT model type, except that the training method is to mask the sentence as a unit. For example, when training BERT, three sentences ABC are input, and the training is no longer just randomly masking the words in the sentence, but masking the entire sentence. For example, all the words in sentence B are known, and all the words in sentences A and C are masked, and then BERT predicts the masked words), BERT outputs the sentence-level Embedding, and then the two Embeddings are respectively passed through two fully connected layers with different parameters, and the vector dimensions of the two Embeddings are mapped to the same dimension. Finally, the two Embedding vectors are pooled using average pooling, and the sentence embedding feature vector corresponding to the input text language is output.

[0018] Multi-modal Processing Module The input of the multimodal processing module has the same dimension The visual representation feature vector and sentence embedding feature vector The multimodal processing module uses vector element addition ( ), vector element multiplication ( ), vector union (∥), and full connection ( ) layer to combine the information from the two feature vectors. The addition and multiplication operations allow the addition and multiplication interaction between the two modal feature vectors without changing the feature dimension. The FC layer allows interaction between all elements. The input dimension of the layer is 2* , the output dimension is The outputs of all three operations are concatenated to construct a multimodal representation feature vector , the multimodal representation feature vector is the input of the temporal localization regression network.

[0019] Temporal Localization Regression Networks The temporal localization regression network represents the feature vector in a multimodal manner As input, there are two same-level output layers. The first output layer outputs the output sentence and video clips The alignment score between The second output layer outputs the position regression offset of the video clip. Furthermore, the temporal localization regression network defines two position offsets, the first of which is a parameterized offset: ,in and They are the parameterized center point offset and length offset respectively. The parameterization is as follows: in and Represents the center coordinates and clip length of the video clip respectively. , for prediction clips and test clips (with The second parameterized offset is a non-parameterized offset: ,in and are the start and end point offsets.

[0020] in and denote the start and end coordinates of the video clip, respectively. The position regression offset can be thought of as a clip position regression from the test clip to a nearby real data clip. Since the original clip may be too tight or too loose, the regression process tends to find a better position.

[0021] Multi-task Loss Function As mentioned above, the temporal localization regression network includes two output layers of the same level, one for alignment and the other for regression. In this application, a multi-task loss L is designed to jointly train visual semantic alignment and clip position regression, which is also called segment position regression. represents visual semantic alignment, represents the fragment position regression, is a hyperparameter that controls the balance of the two task losses. The alignment loss will cause aligned fragment-sentence pairs to get positive scores, while misaligned fragment-sentence pairs will get negative scores.

[0022] N is the batch size, For Sentences With video clips The alignment score between and is a parameter that controls the weight between positive (aligned) and negative (misaligned) clip-sentence pairs.

[0023] In this application, the temporal activity localization model aligns the clipped sentence pairs to calculate the regression loss . A sentence The annotation contains the start and end time ( ). Aligned sliding window clip have( . where x and y represent the parameterized offset and , s and e represent non-parametric offsets. is smooth Function. Actual data offset Calculated from the start and end time.

[0024] In one embodiment, if Figure 3 As shown, the temporal activity positioning model is trained in the following way: S101, using the visual encoder of the temporal activity localization model, transforming each of the marketing video clips into a visual representation feature vector; S102, using the sentence encoder of the temporal activity positioning model, mapping the sentence description embedding code corresponding to each marketing video clip into a sentence embedding feature vector; S103, inputting each corresponding visual representation feature vector and the sentence embedding feature vector into the multimodal processing module of the temporal activity localization model to obtain a multimodal representation feature vector; S104. Input the multimodal representation feature vector of each sample insurance product marketing video into the temporal localization regression network of the temporal activity localization model, and perform weighted training with a visual semantic alignment loss function and a fragment position regression function to obtain a temporal activity localization model that meets the conditions.

[0025] For the corresponding processing procedures of the training process, please refer to the corresponding description of the processing procedures of the time activity positioning model mentioned above. The difference lies in that the training data of the embodiment of the present application includes multiple sample insurance product marketing videos, each of the sample insurance product marketing videos includes a marketing video clip and a sentence description of the marketing video clip, and the sentence description is used to characterize the start and end sentences of the marketing video clip. The marketing video clips of the multiple sample insurance product marketing videos correspond to different insurance consultation types, and the specific description will not be repeated here.

[0026] In this embodiment, by collecting sample insurance product marketing videos and training the time activity positioning model, the final time activity positioning model used in this case is obtained, which is used to output marketing video clips corresponding to the insurance consulting text language in the actual insurance product marketing video.

[0027] It should be noted that in order to collect training sample data, a multi-scale time sliding window with [64, 128, 256, 512] frames and 80% overlap was used to sample and edit different sample insurance product marketing videos to obtain multiple sample data.

[0028] In addition, in this embodiment, when collecting sample data, the IoU (Intersection over Union) and nIoL (non-Intersection over Length) indicators can be used to measure or determine whether the data can be used for training. IoU and nIoL are two indicators used to measure the degree of overlap between two time periods in a sample insurance product marketing segment. represents time period a, Indicates time period b. Indicates the length of the intersection of two time periods. Indicates the length of the union of two time periods, Indicates the length of the portion of time A that does not overlap with time B.

[0029] When the two time periods overlap completely, IoU is 1 and nIoL is 0. When the two time periods do not overlap at all, IoU is 0 and nIoL is 1. If the IoU value is high, the nIoL value will usually be low because there is more overlap between the two time periods.

[0030] Training samples Each training sample contains a sentence description and a video clip For the annotations from C The sliding window clip c and has time annotations The sentence description s, in this embodiment, if it satisfies: (1) When the IoU (i.e., the ratio of the overlapping part of the temporal annotations to the total time period) is greater than a first preset threshold (e.g., 0.5), the corresponding two marketing video clips are aligned as a pair of training samples; (2) nIoL (i.e., the ratio of the non-overlapping part of the temporal annotation to the length of the sentence description) must be less than a second preset threshold (e.g., 0.2); (3) Each sliding window clip can only be paired with one sentence description.

[0031] In this embodiment, the reason for using nIoL is that it is hoped that most of the content of the sample video clips is associated with the sentence description, thereby improving the effectiveness and speed of training.

[0032] In one embodiment, the sample insurance product marketing video is obtained by: S201, determining a plurality of different product keywords of the same insurance product from an insurance product knowledge base; S202, searching for a plurality of insurance product marketing videos matching the product keyword from the insurance product marketing video library to obtain an insurance product marketing video set; S203, annotating each insurance product marketing video in the insurance product marketing video set with at least one insurance activity category time to obtain a sentence description of each insurance product marketing video, wherein the insurance activity category includes equity activities; S204: Use the insurance product marketing videos in the insurance product marketing video set that have been annotated with insurance activity categories and times as sample insurance product marketing videos.

[0033] Among them, the insurance activity categories include but are not limited to equity activities, product type introduction, product cost introduction, insurance product enterprise introduction, insurance product cooperative bank introduction, insurance product term and other activity categories. In this embodiment, when collecting sample insurance product marketing videos, multiple different product keywords (such as product name, type and other keywords) of the same insurance product (such as insurance product A) are determined from the insurance product knowledge base, and then multiple insurance product marketing videos matching the product keywords are directly searched from the insurance product marketing video library to obtain an insurance product marketing video set, which can first quickly collect insurance product marketing videos of the same insurance product; then, at least one insurance activity category time annotation is performed on each insurance product marketing video in the insurance product marketing video set, and this process can be changed according to actual needs and is not specifically limited. Then, the insurance product marketing videos in the insurance product marketing video set that have been annotated with insurance activity category time are used as sample insurance product marketing videos to quickly find sample insurance product marketing videos of the required insurance product type, and implement customized activity category annotations to adapt to actual consulting situations.

[0034] In one embodiment, if Figure 4 As shown, based on the temporal activity positioning model provided in the embodiment of the present application, the embodiment of the present application provides a video processing method, comprising the following steps: S10: Acquire the insurance consultation text language input by the user, where the insurance consultation text language is input by the user for the insurance product marketing video; S20: Inputting the insurance consulting text language and the insurance product marketing video into a pre-trained time activity positioning model to cut out a marketing video clip corresponding to the insurance consulting text language from the insurance product marketing video; The trained time activity positioning model is obtained by training based on training sample data, and the training sample data includes multiple sample insurance product marketing videos, each of which includes a marketing video clip and a sentence description of the marketing video clip, and the sentence description is used to characterize the start and end sentences of the marketing video clip. The marketing video clips of the multiple sample insurance product marketing videos correspond to different insurance consultation types. Please refer to the above description for the specific training process, which will not be described in detail here.

[0035] In this embodiment, after promoting or pushing an insurance product marketing video to a user, the insurance consultation text language input by the user can be obtained, such as insurance consultation text language input manually or by voice. The insurance consultation text language is input by the user for the insurance product marketing video. For example, the insurance consultation text language includes consultation on one or more of the following situations: rights and interests activities, introduction to product categories, introduction to product costs, introduction to insurance product companies, introduction to insurance product cooperative banks, insurance product terms, etc., without specific limitation. The insurance consultation text language and the insurance product marketing video are then input into a pre-trained time activity positioning model to cut out a marketing video clip corresponding to the insurance consultation text language from the insurance product marketing video. For example, the consultation text language for consulting on rights and interests activities is input, and then the marketing video clip corresponding to the rights and interests activities is processed and cut out from the insurance product marketing video.

[0036] It can be seen that in this embodiment, an artificial intelligence-based video processing method is provided, which obtains insurance consulting text language input by a user, wherein the insurance consulting text language is input by the user for an insurance product marketing video, and the insurance consulting text language and the insurance product marketing video are input into a pre-trained time-activity positioning model to cut out a marketing video clip corresponding to the insurance consulting text language from the insurance product marketing video. In this way, users can quickly locate the video content part they want from the insurance product marketing video, thereby improving the low efficiency of locating the desired video clip of the insurance product marketing video, thereby improving the user experience and the marketing conversion rate of insurance products.

[0037] In one embodiment, if Figure 5 As shown, after step S20, after the method inputs the insurance consultation text language and the insurance product marketing video into the pre-trained time activity positioning model, the method further includes the following steps: S30: Determine whether the model successfully outputs a marketing video clip corresponding to the insurance consulting text language cut out from the insurance product marketing video; S40: When the time-activity positioning model cannot cut out a marketing video segment corresponding to the insurance consulting text language from the insurance product marketing video, determining a video keyword of the insurance product marketing video; S50: Matching the video keyword with a keyword in a preset prompt text library to match a target prompt text corresponding to the video keyword, wherein the preset prompt text library contains a plurality of prompt texts corresponding to the insurance product; S60: Output a prompt pop-up box, wherein the content of the prompt pop-up box includes the target prompt text. In this embodiment, after the insurance consultation text language and the insurance product marketing video are input into the pre-trained time activity positioning model, there are two situations, one is that the corresponding marketing video clip is successfully output, and the other is that the time activity positioning model cannot cut out the marketing video clip corresponding to the insurance consultation text language from the insurance product marketing video. When the time activity positioning model cannot cut out the marketing video clip corresponding to the insurance consultation text language from the insurance product marketing video, the video keyword of the insurance product marketing video is determined. Video keywords, such as insurance product name, content keywords, etc., such as rights and interests type, can be extracted through video parsing technology. Then, the video keywords are matched with keywords in the preset prompt text library to match the target prompt text corresponding to the video keywords, wherein the preset prompt text library is a text library that is continuously updated and maintained, and the preset prompt text library contains multiple prompt texts corresponding to the insurance product, and the contents of the multiple prompt texts are constructed based on the insurance product marketing video corresponding to the insurance product.

[0038] For example, if insurance product A is parsed out from the keywords of the video, it means that the model cannot output the content related to insurance product A from the insurance product marketing video based on the user input language. Therefore, some prompt texts in the insurance product A are found from the preset prompt text library and output to the prompt pop-up box, thereby reminding the user to convert other prompt texts related to the insurance product marketing video content of insurance product A, such as prompting the user to consult other aspects of the insurance product A, thereby assisting the user to input the locateable consultation language, ensuring the interactive continuity of the user's consultation of the insurance product marketing video, and avoiding the embarrassing situation of no result output suddenly, which greatly improves the user experience.

[0039] In one embodiment, step S50, that is, matching the video keyword with the keyword in the preset prompt text library to match the target prompt text corresponding to the video keyword, includes the following steps: S51: Obtaining the insurance product type corresponding to the insurance product marketing video; S52: Determine a preset prompt text library corresponding to the insurance product type; S53: Matching the video keyword with a keyword in a preset prompt text library corresponding to the insurance product type to match a target prompt text corresponding to the video keyword.

[0040] In this embodiment, in order to adapt to the tailoring of various insurance product types, in this embodiment, a plurality of preset prompt text libraries corresponding to the different insurance product types are set. In the process of matching the video keywords with the keywords in the preset prompt text library to match the target prompt text corresponding to the video keywords, the insurance product type corresponding to the insurance product marketing video, such as insurance product A, is obtained, thereby determining the preset prompt text library corresponding to the insurance product A, and matching the video keywords with the keywords in the preset prompt text library corresponding to the insurance product type to match the target prompt text corresponding to the video keywords.

[0041] In this embodiment, it is possible to ensure that the target prompt text corresponding to the video keyword is matched to the maximum extent, thereby improving the interactive experience. In other embodiments, it is also possible to match from the prompt text library of other insurance types, which can reduce the maintenance work of the library and also ensure the output of effective prompt text. Because different insurance product types are different, but the main content is mostly rights, prices, cooperative banks and other main contents.

[0042] In one embodiment, after step S20, after inputting the insurance consultation text language and the insurance product marketing video into a pre-trained time activity positioning model, the method further includes the following steps: S70: When the time activity positioning model cannot cut out the marketing video clip corresponding to the insurance consulting text language from the insurance product marketing video, parse out the text keywords of the insurance consulting text language; S80: Searching for insurance product marketing videos matching the text keyword from a preset insurance product marketing video library; S90: Input the insurance consulting text language and the matching insurance product marketing video into the time activity positioning model to cut out the marketing video clip corresponding to the insurance consulting text language from the matching insurance product marketing video.

[0043] In this embodiment, when the time activity positioning model cannot cut out the marketing video clip corresponding to the insurance consultation text language from the insurance product marketing video, the text keywords of the insurance consultation text language are parsed, such as the content part that is not involved in the consultation video, and the text keywords of the insurance consultation text language are parsed to see whether the insurance product marketing video matching the text keywords can be searched from the preset insurance product marketing video library. If so, the insurance consultation text language and the matching insurance product marketing video are input into the time activity positioning model to cut out the marketing video clip corresponding to the insurance consultation text language from the matching insurance product marketing video. In other words, this embodiment can also maintain a preset insurance product marketing video library, and the videos in the preset insurance product marketing video library are associated with the content of the insurance product marketing video. In order to prevent the corresponding video clip from being unable to be located from video 1, the video can be found in the library for positioning, thereby ensuring the continuity of user interaction.

[0044] In one embodiment, after step S20, that is, after the marketing video clip corresponding to the insurance consultation text language is cut out from the insurance product marketing video, the following steps are also included: S100: associating the text keywords of the insurance consultation text language with the cut marketing video clips and storing them in a consultation video clip library; S110: When the insurance consultation text language is received again, a marketing video clip corresponding to the insurance consultation text language is matched and outputted from the consultation video clip library.

[0045] In this embodiment, as another processing mechanism, a marketing video clip corresponding to the insurance consulting text language is cut out from the insurance product marketing video, and the text keywords of the insurance consulting text language and the cut marketing video clip are associated and stored in the consulting video clip library. When the insurance consulting text language is received again, the marketing video clip corresponding to the insurance consulting text language is matched and output from the consulting video clip library. In this way, multiple calls to model processing can be greatly avoided or reduced, and pre-stored videos can also be quickly rendered for display, thereby improving processing efficiency.

[0046] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0047] In one embodiment, a video processing device based on artificial intelligence is provided, and the video processing device based on artificial intelligence corresponds one-to-one with the video processing method based on artificial intelligence in the above embodiment. Figure 6As shown, the video processing device based on artificial intelligence includes an acquisition module 101 and an input module 102. Each functional module is described in detail as follows: The acquisition module 101 is used to acquire the insurance consultation text language input by the user, wherein the insurance consultation text language is input by the user for the insurance product marketing video; An input module 102 is used to input the insurance consulting text language and the insurance product marketing video into a pre-trained time activity positioning model to cut out a marketing video clip corresponding to the insurance consulting text language from the insurance product marketing video; Among them, the trained time-activity positioning model is trained based on training sample data, and the training sample data includes multiple sample insurance product marketing videos, each of the sample insurance product marketing videos includes a marketing video clip and a sentence description of the marketing video clip, and the sentence description is used to characterize the start and end sentences of the marketing video clip, and the marketing video clips of the multiple sample insurance product marketing videos correspond to different insurance consultation types. In one embodiment, after the insurance consultation text language and the insurance product marketing video are input into the pre-trained time activity positioning model, the video processing device is further used to: When the time activity positioning model cannot cut out the marketing video clip corresponding to the insurance consultation text language from the insurance product marketing video, determining the video keywords of the insurance product marketing video; Matching the video keyword with a keyword in a preset prompt text library to match a target prompt text corresponding to the video keyword, wherein the preset prompt text library contains a plurality of prompt texts corresponding to the insurance product; A prompt pop-up box is output, wherein the content of the prompt pop-up box includes the target prompt text. In one embodiment, the video processing device is further used for: Obtain the insurance product type corresponding to the insurance product marketing video; Determine a preset prompt text library corresponding to the insurance product type; The video keywords are matched with keywords in a preset prompt text library corresponding to the insurance product type to match target prompt texts corresponding to the video keywords.

[0048] In one embodiment, after the insurance consultation text language and the insurance product marketing video are input into the pre-trained time activity positioning model, the video processing device is further used to: When the time activity positioning model cannot cut out the marketing video clip corresponding to the insurance consulting text language from the insurance product marketing video, parse out the text keywords of the insurance consulting text language; Searching for insurance product marketing videos matching the text keyword from a preset insurance product marketing video library; The insurance consulting text language and the matching insurance product marketing video are input into the time activity positioning model to cut out the marketing video clip corresponding to the insurance consulting text language from the matching insurance product marketing video.

[0049] In one embodiment, the temporal activity localization model is trained in the following manner: transforming each of the marketing video clips into a visual representation feature vector using a visual encoder of the temporal activity localization model; Using the sentence encoder of the temporal activity positioning model, the sentence description embedding code corresponding to each marketing video clip is mapped into a sentence embedding feature vector; Input each corresponding visual representation feature vector and sentence embedding feature vector into the multimodal processing module of the temporal activity localization model to obtain a multimodal representation feature vector; The multimodal representation feature vector of each sample insurance product marketing video is input into the temporal localization regression network of the temporal activity localization model, and weighted training is performed using a visual semantic alignment loss function and a fragment position regression function to obtain a temporal activity localization model that meets the conditions.

[0050] In one embodiment, the sample insurance product marketing video is obtained by: Determine a plurality of different product keywords for the same insurance product from an insurance product knowledge base; Searching for multiple insurance product marketing videos matching the product keyword from the insurance product marketing video library to obtain an insurance product marketing video set; Performing time annotation of at least one insurance activity category on each insurance product marketing video in the insurance product marketing video set to obtain a sentence description of each insurance product marketing video, wherein the insurance activity category includes equity activities; The insurance product marketing videos in the insurance product marketing video set that have been annotated with insurance activity categories and times are used as sample insurance product marketing videos.

[0051] In one embodiment, after cutting out the marketing video clip corresponding to the insurance consultation text language from the insurance product marketing video, the video processing device is further used to: Associating the text keywords of the insurance consultation text language with the cut marketing video clips and storing them in a consultation video clip library; When the insurance consultation text language is received again, a marketing video clip corresponding to the insurance consultation text language is matched and outputted from the consultation video clip library.

[0052] It can be seen that in this embodiment, an artificial intelligence-based video processing device is provided, which obtains insurance consulting text language input by a user, wherein the insurance consulting text language is input by the user for an insurance product marketing video, and the insurance consulting text language and the insurance product marketing video are input into a pre-trained time-activity positioning model to cut out a marketing video clip corresponding to the insurance consulting text language from the insurance product marketing video. In this way, the user can quickly locate the video content part he wants from the insurance product marketing video, thereby improving the low efficiency of locating the desired video clip of the insurance product marketing video, thereby improving the user experience and the marketing conversion rate of the insurance product. For the specific definition of the video processing device based on artificial intelligence, please refer to the definition of the video processing method based on artificial intelligence above, which will not be repeated here. Each module in the above-mentioned video processing device based on artificial intelligence can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0053] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a video processing method based on artificial intelligence is implemented.

[0054] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a video processing method based on artificial intelligence in the above embodiment is implemented, for example Figure 4Alternatively, when the processor executes the computer program, the functions of each module / unit in the embodiment of the video processing device based on artificial intelligence are realized, which will not be described here in order to avoid repetition.

[0055] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, an artificial intelligence-based video processing method in the above embodiment is implemented, for example Figure 4 Alternatively, when the processor executes the computer program, the functions of each module / unit in the embodiment of the video processing device based on artificial intelligence are realized, which will not be described here in order to avoid repetition.

[0056] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0057] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0058] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A video processing method based on artificial intelligence, characterized in that: include: Acquire the insurance consultation text language input by the user, where the insurance consultation text language is input by the user for the insurance product marketing video; Inputting the insurance consulting text language and the insurance product marketing video into a pre-trained time activity positioning model to cut out marketing video clips corresponding to the insurance consulting text language from the insurance product marketing video; Among them, the trained time-activity positioning model is trained based on training sample data, and the training sample data includes multiple sample insurance product marketing videos, each of the sample insurance product marketing videos includes a marketing video clip and a sentence description of the marketing video clip, and the sentence description is used to characterize the start and end sentences of the marketing video clip, and the marketing video clips of the multiple sample insurance product marketing videos correspond to different insurance consultation types.

2. The video processing method based on artificial intelligence according to claim 1, characterized in that: After inputting the insurance consultation text language and the insurance product marketing video into a pre-trained time activity positioning model, the method further includes: When the time activity positioning model cannot cut out the marketing video clip corresponding to the insurance consultation text language from the insurance product marketing video, determining the video keywords of the insurance product marketing video; Matching the video keyword with a keyword in a preset prompt text library to match a target prompt text corresponding to the video keyword, wherein the preset prompt text library contains a plurality of prompt texts corresponding to the insurance product; A prompt pop-up box is output, wherein the content of the prompt pop-up box includes the target prompt text.

3. The video processing method based on artificial intelligence according to claim 2, characterized in that: The step of matching the video keyword with a keyword in a preset prompt text library to obtain a target prompt text corresponding to the video keyword includes: Obtain the insurance product type corresponding to the insurance product marketing video; Determine a preset prompt text library corresponding to the insurance product type; The video keywords are matched with keywords in a preset prompt text library corresponding to the insurance product type to match target prompt texts corresponding to the video keywords.

4. The video processing method based on artificial intelligence according to claim 1, characterized in that: After inputting the insurance consultation text language and the insurance product marketing video into a pre-trained time activity positioning model, the method further includes: When the time activity positioning model cannot cut out the marketing video clip corresponding to the insurance consulting text language from the insurance product marketing video, parse out the text keywords of the insurance consulting text language; Searching for insurance product marketing videos matching the text keyword from a preset insurance product marketing video library; The insurance consulting text language and the matching insurance product marketing video are input into the time activity positioning model to cut out the marketing video clip corresponding to the insurance consulting text language from the matching insurance product marketing video.

5. The video processing method based on artificial intelligence according to any one of claims 1 to 4, characterized in that: The temporal activity localization model is trained in the following way: transforming each of the marketing video clips into a visual representation feature vector using a visual encoder of the temporal activity localization model; Using the sentence encoder of the temporal activity positioning model, the sentence description embedding code corresponding to each marketing video clip is mapped into a sentence embedding feature vector; Input each corresponding visual representation feature vector and sentence embedding feature vector into the multimodal processing module of the temporal activity localization model to obtain a multimodal representation feature vector; The multimodal representation feature vector of each sample insurance product marketing video is input into the temporal localization regression network of the temporal activity localization model, and weighted training is performed using a visual semantic alignment loss function and a fragment position regression function to obtain a temporal activity localization model that meets the conditions.

6. The video processing method based on artificial intelligence according to claim 5, characterized in that: The sample insurance product marketing video is obtained in the following way: Determine a plurality of different product keywords for the same insurance product from an insurance product knowledge base; Searching for multiple insurance product marketing videos matching the product keyword from the insurance product marketing video library to obtain an insurance product marketing video set; Performing time annotation of at least one insurance activity category on each insurance product marketing video in the insurance product marketing video set to obtain a sentence description of each insurance product marketing video, wherein the insurance activity category includes equity activities; The insurance product marketing videos in the insurance product marketing video set that have been annotated with insurance activity categories and times are used as sample insurance product marketing videos.

7. The video processing method based on artificial intelligence according to any one of claims 1 to 4, characterized in that: After cutting out the marketing video clip corresponding to the insurance consultation text language from the insurance product marketing video, the method further includes: Associating the text keywords of the insurance consultation text language with the cut marketing video clips and storing them in a consultation video clip library; When the insurance consultation text language is received again, a marketing video clip corresponding to the insurance consultation text language is matched and outputted from the consultation video clip library.

8. A video processing device based on artificial intelligence, characterized in that: include: An acquisition module, used to acquire the insurance consultation text language input by the user, wherein the insurance consultation text language is input by the user for the insurance product marketing video; An input module, used to input the insurance consulting text language and the insurance product marketing video into a pre-trained time activity positioning model, so as to cut out the marketing video clip corresponding to the insurance consulting text language from the insurance product marketing video; Among them, the trained time-activity positioning model is trained based on training sample data, and the training sample data includes multiple sample insurance product marketing videos, each of the sample insurance product marketing videos includes a marketing video clip and a sentence description of the marketing video clip, and the sentence description is used to characterize the start and end sentences of the marketing video clip, and the marketing video clips of the multiple sample insurance product marketing videos correspond to different insurance consultation types.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the artificial intelligence-based video processing method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the artificial intelligence-based video processing method according to any one of claims 1 to 7 is implemented.

Images