Training processing method of video search model and video search processing method and device

Abstract

The embodiment of the invention relates to a training processing method of a video search model and a video search processing method and device, and the method comprises the steps: obtaining a sample video retrieval text, and extracting a first sample retrieval text feature of the sample video retrieval text through an initial video search model; distributing a plurality of first sample feature weights in one-to-one correspondence with preset multi-mode video information; determining a first sample video matched with the first sample retrieval text feature and the plurality of first sample feature weights through an initial video search model; and calculating a first matching degree between the first sample video and the sample video retrieval text, and when the first matching degree is smaller than a preset matching degree threshold value, adjusting model parameters of the initial video search model until the adjusted matching degree is smaller than the preset matching degree threshold value, thereby obtaining a trained video search model. According to the technical scheme, the matching degree of the searched video clip and the video retrieval text is ensured, and the video retrieval effect is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer application technology, and in particular to a video search model training processing method, a video search processing method and a device. Background Art

[0002] With the rapid development of the Internet and multimedia technology, the speed of production and consumption of various types of videos such as movies, TV series, news reports, and educational tutorials has greatly accelerated. This requires an effective video retrieval technology so that users can quickly find the required video clips.

[0003] In related technologies, video retrieval mainly relies on the search text input by the user, and matches the text features with the multimodal features of the video clip to be retrieved (such as video screen features, audio features, and dialogue features). For example, the similarity between the text features and each modal feature in the time to be retrieved is calculated, and the video clips with high matching degree are screened out based on the average of multiple similarities corresponding to the multimodal features.

[0004] However, in the above video retrieval method, the text features are directly compared with the multimodal features corresponding to the multimodal video information for feature similarity, without considering the differences between multiple modal features, resulting in a low match between the retrieved video clips and the video retrieval text. Summary of the invention

[0005] In order to solve the above technical problems or at least partially solve the above technical problems, the present disclosure provides a video search model training processing method, a video search processing method and a device.

[0006] The disclosed embodiment provides a training and processing method for a video search model, comprising the following steps: obtaining a sample video retrieval text, and inputting the sample video retrieval text into a pre-constructed initial video search model, extracting a first sample retrieval text feature of the sample video retrieval text through the initial video search model; allocating a plurality of first sample feature weights corresponding to preset multimodal video information according to the first sample retrieval text feature through the initial video search model; determining a first sample video that matches the first sample retrieval text feature and the plurality of first sample feature weights according to a preset video feature library through the initial video search model; calculating a first matching degree between the first sample video and the sample video retrieval text, and determining whether the first matching degree is less than a preset matching degree threshold; and if the first matching degree is less than the preset matching degree threshold, adjusting the model parameters of the initial video search model until the adjusted matching degree is less than the preset matching degree threshold to obtain a trained video search model.

[0007] Optionally, the model parameters of the initial video search model are adjusted until the adjusted matching degree is less than the preset matching degree threshold to obtain a trained video search model, including: after adjusting the model parameters of the initial video search model, the sample video retrieval text is input into the adjusted video search model, and a second sample retrieval text feature of the sample video retrieval text is extracted through the adjusted video search model; a plurality of second sample feature weights corresponding one-to-one to the multimodal video information are assigned according to the second sample retrieval text feature through the adjusted video search model; a second sample video matching the second sample retrieval text feature and the plurality of second sample feature weights is determined according to the preset video feature library through the adjusted video search model; a second matching degree between the second sample video and the sample video retrieval text is calculated to determine whether the second matching degree is less than the preset matching degree threshold; when it is not less than the preset matching degree threshold, determining that the adjusted video search model is the trained video search model.

[0008] Optionally, after determining whether the second matching degree is less than the preset matching degree threshold, it also includes: when it is less than the preset matching degree threshold, adjusting the model parameters of the adjusted video search model until the trained video search model is obtained when the adjusted matching degree is less than the preset matching degree threshold.

[0009] Optionally, the preset video feature library includes multiple video feature sets corresponding one-to-one to multiple sample videos, and each of the video feature sets includes multiple video features corresponding one-to-one to the corresponding sample video and the multimodal video information.

[0010] Optionally, the method of determining the first sample video that matches the first sample retrieval text feature and the multiple first sample feature weights through the initial video search model according to a preset video feature library includes: calculating the feature similarity between the first sample retrieval text feature and each of the video features in each of the video feature sets; calculating the product value of each of the feature similarities and the corresponding first sample feature weight; calculating the sum of all the product values ​​corresponding to each of the video feature sets; and determining the matching first sample video according to the sum value.

[0011] Optionally, determining the matching first sample video according to the sum value includes: sorting all the sum values ​​corresponding to all the video feature sets in order from high to low to obtain a sorting result; and determining the sample video corresponding to the first sum value in the sorting result as the matching first sample video.

[0012] Optionally, the calculation of the first matching degree between the first sample video and the sample video retrieval text includes: calculating the loss value of the first sample video and the sample video retrieval text respectively according to each type of loss function in a preset multi-category loss function to obtain multiple loss values ​​corresponding to the multi-category loss function; and calculating the average of the multiple loss values ​​to obtain the first matching degree.

[0013] Optionally, the multimodal video information includes multiple of the following: video picture information, audio information, video text description information, dialogue subtitle information, and video evaluation information.

[0014] An embodiment of the present disclosure provides a video search processing method, comprising the following steps: in response to obtaining a video retrieval instruction carrying a video retrieval text, inputting the video retrieval text into a pre-trained video search model; determining a target video matching the video retrieval text through the video search model according to a preset video feature library, wherein the video search model is obtained according to the aforementioned video search model training processing method.

[0015] Optionally, determining a target video that matches the video retrieval text through the video search model according to a preset video feature library includes: extracting retrieval text features corresponding to the video retrieval text through the video search model, and determining multiple feature weights that correspond one-to-one to preset multimodal video information based on the retrieval text features; determining the matching target video in the preset video feature library based on the retrieval text features and the multiple feature weights through the video search model.

[0016] The disclosed embodiment also provides a training and processing device for a video search model, comprising: an acquisition module for acquiring a sample video retrieval text; an extraction module for inputting the sample video retrieval text into a pre-constructed initial video search model, and extracting a first sample retrieval text feature of the sample video retrieval text through the initial video search model; an allocation module for allocating a plurality of first sample feature weights corresponding to preset multimodal video information according to the first sample retrieval text feature through the initial video search model; a first determination module for determining a first sample video that matches the first sample retrieval text feature and the plurality of first sample feature weights through the initial video search model according to a preset video feature library; a second determination module for calculating a first matching degree between the first sample video and the sample video retrieval text, and determining whether the first matching degree is less than a preset matching degree threshold; and a training processing module for adjusting the model parameters of the initial video search model when the matching degree is less than the preset matching degree threshold, until a trained video search model is obtained when the adjusted matching degree is less than the preset matching degree threshold.

[0017] The disclosed embodiment also provides a video search processing device, including: an output module, for responding to obtaining a video retrieval instruction carrying a video retrieval text, and inputting the video retrieval text into a pre-trained video search model; a third determination module, for determining a target video matching the video retrieval text through the video search model according to a preset video feature library, wherein the video search model is obtained according to the aforementioned video search model training processing method.

[0018] An embodiment of the present disclosure also provides an electronic device, which includes: a processor; a memory for storing executable instructions of the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement a training processing method for a video search model or a video search processing method as provided in an embodiment of the present disclosure.

[0019] The embodiments of the present disclosure also provide a computer-readable storage medium, which stores a computer program, and the computer program is used to execute the video search model training processing method or the video search processing method provided by the embodiments of the present disclosure.

[0020] Compared with the prior art, the technical solution provided by the embodiments of the present disclosure has the following advantages:

[0021] The video search processing scheme provided by the disclosed embodiment obtains a sample video retrieval text, and inputs the sample video retrieval text into a pre-built initial video search model, extracts the first sample retrieval text feature of the sample video retrieval text through the initial video search model, and assigns multiple first sample feature weights corresponding to preset multimodal video information according to the first sample retrieval text feature through the initial video search model, and then, through the initial video search model, according to the preset video feature library, determines the first sample video that matches the first sample retrieval text feature and multiple first sample feature weights, calculates the first matching degree between the first sample video and the sample video retrieval text, determines whether the first matching degree is less than a preset matching degree threshold, and when it is less than the preset matching degree threshold, adjusts the model parameters of the initial video search model until the adjusted matching degree is less than the preset matching degree threshold to obtain a trained video search model. In this technical solution, the matching degree between the searched video clip and the video retrieval text is guaranteed, and the video retrieval effect is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.

[0023] Figure 1 A schematic diagram of a flow chart of a video search processing method provided by an embodiment of the present disclosure;

[0024] Figure 2 A flowchart of another video search processing method provided by an embodiment of the present disclosure;

[0025] Figure 3 A schematic diagram of the structure of a video search processing device provided by an embodiment of the present disclosure;

[0026] Figure 4 A schematic diagram of the structure of another video search processing device provided by an embodiment of the present disclosure;

[0027] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0028] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0029] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0030] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0031] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0032] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0033] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0034] As mentioned above, the differences between multimodal features are not taken into account during video retrieval, which may result in low accuracy of retrieved video clips. The applicant found that in actual applications, the video retrieval text is different, and the modal features with high contribution in the multiple feature weights corresponding to the multimodal feature video information are different. For example, for the video retrieval text "The male protagonist takes a swimsuit photo at the sunset on the beach", the contribution of the video picture features is higher. For example, for the video retrieval text "The male protagonist says to the female protagonist: "I love you"", the contribution of the lines features is higher.

[0035] Therefore, in order to solve the above technical problems, the present disclosure proposes a solution for dynamically determining the feature weights of multimodal video information based on video retrieval text. In this solution, the video retrieval text is adapted to dynamically determine multiple feature weights corresponding to the multimodal video information, thereby ensuring the accuracy of video retrieval.

[0036] Specifically, the embodiment of the present disclosure provides a video search model training processing method, which can be performed by a video search model training processing device, wherein the device can be implemented by software and / or hardware, and can generally be integrated in an electronic device. Figure 1 As shown, the method includes:

[0037] Step 101, obtain a sample video retrieval text, and input the sample video retrieval text into a pre-built initial video search model, and extract a first sample retrieval text feature of the sample video retrieval text through the initial video search model.

[0038] In one embodiment of the present disclosure, a sample video retrieval text is obtained, and the sample video retrieval text is input into a pre-built initial video search model, and a first sample retrieval text feature of the sample video retrieval text is extracted by the initial video search model. The first sample retrieval text feature can be extracted by a feature extraction layer of the initial video search model, and the specific extraction method is related to the network structure of the initial video search model, etc.

[0039] Step 102: allocating a plurality of first sample feature weights corresponding one-to-one to the preset multimodal video information according to the first sample retrieval text feature through an initial video search model.

[0040] Among them, the multi-video modal information includes multiple of the following: video picture information, audio information, video text description information, dialogue subtitle information, video evaluation information, background music, etc.

[0041] It should be noted that the initial video retrieval model includes a weight distribution module, which is used to assign multiple first sample feature weights corresponding to the preset multimodal video information based on the first sample retrieval text. Therefore, in this embodiment, the initial video search model assigns multiple first sample feature weights corresponding to the preset multimodal video information based on the first sample retrieval text features. The multiple first sample feature weights are usually not exactly the same.

[0042] Step 103: Determine, through an initial video search model and according to a preset video feature library, a first sample video that matches the first sample retrieval text feature and a plurality of first sample feature weights.

[0043] The preset video feature library is a pre-constructed set of multiple video features corresponding to multiple sample videos, each of which includes multiple video features corresponding to the corresponding sample video and multimodal video information. The multiple video clip features can be pre-extracted by a pre-trained convolutional neural network.

[0044] In one embodiment of the present disclosure, the feature similarity between the first sample retrieval text feature and each video feature in each video feature set is calculated, wherein the cosine distance between the first sample retrieval text feature and each video feature in each video feature set can be calculated, and the cosine distance is used as the corresponding feature similarity, etc. The product value of each feature similarity and the corresponding first sample feature weight is calculated, the sum value of all product values ​​corresponding to each video feature set is calculated, and the matching first sample video is determined according to the sum value.

[0045] In this embodiment, all sum values ​​corresponding to all video feature sets may be sorted in descending order to obtain a sorting result, and the sample video corresponding to the first sum value in the sorting result is determined as the matching first sample video.

[0046] Step 104, calculating a first matching degree between the first sample video and the sample video search text, and determining whether the first matching degree is less than a preset matching degree threshold.

[0047] In one embodiment of the present disclosure, a first matching degree between a first sample video and a sample video retrieval text is calculated to determine whether the first matching degree is less than a preset matching degree threshold, wherein the preset matching degree threshold can be calibrated according to an experimental scenario.

[0048] It should be noted that in different application scenarios, the methods for calculating the first matching degree between the first sample video and the sample video search text are different, as shown in the following examples:

[0049] In some possible embodiments, sample video features of the first sample video may be extracted, sample text features of the sample search text may be extracted, a cosine distance between the sample video features and the sample search text features may be calculated, and the cosine distance may be determined as a matching degree.

[0050] In some possible embodiments, standard sample videos that match the sample retrieval text may be pre-marked. After obtaining the first sample video, the video matching degree between the first sample video and the standard sample video is directly calculated, and the video matching degree is determined to be the matching degree between the first sample video and the sample retrieval text.

[0051] Among them, when calculating the first matching degree, in order to ensure the robustness of the calculated first matching degree, the loss value of the first sample video and the sample video retrieval text can be calculated respectively according to each type of loss function in the preset multi-type loss function to obtain multiple loss values ​​corresponding to the multi-type loss function, and the average of the multiple loss values ​​is calculated to obtain the first matching degree. That is, in this embodiment, the first matching degree is calculated in combination with the multi-type loss function, which ensures the reliability of the first matching degree.

[0052] In one embodiment of the present disclosure, multiple types of loss functions may be pre-set, including a preset type of loss function. First, a reference first matching degree is calculated according to the preset type of loss function. After the reference first matching degree is less than a preset matching degree threshold, multiple loss values ​​are calculated based on other types of loss functions. The average of the multiple loss values ​​is calculated as the first matching degree finally determined. When the reference first matching degree is not less than the preset matching degree threshold, the reference first matching degree is directly used as the first matching degree finally determined. This takes into account both the reliability and calculation efficiency of the matching degree calculation.

[0053] Step 105, when the matching degree is less than a preset matching degree threshold, the model parameters of the initial video search model are adjusted until the adjusted matching degree is less than the preset matching degree threshold to obtain a trained video search model.

[0054] After the first matching degree is less than the preset matching degree threshold, the model parameters of the initial video search model are adjusted until the adjusted matching degree is less than the preset matching degree threshold to obtain a trained video search model.

[0055] Among them, in this embodiment, the initial retrieval model can be divided into two modules, one module is the weight distribution module mentioned above, and the other module is a video matching module, wherein the video matching module is used to determine the first sample video that matches the first sample retrieval text feature and multiple first sample feature weights through a preset video feature library.

[0056] In one embodiment of the present disclosure, the video matching module can be matched separately first, that is, a video retrieval module that only includes the video matching module is pre-set. After the video retrieval model is trained, the model parameters of the trained video matching module are reused to construct the above-mentioned initial video retrieval module including the video matching module and the weight distribution module.

[0057] In this embodiment, since the model parameters of the video matching module have been trained, after the first matching degree is less than the preset matching degree threshold, the parameters of the weight distribution module of the initial video search model are adjusted until the trained video search model is obtained when the adjusted matching degree is less than the preset matching degree threshold.

[0058] In one embodiment of the present disclosure, the model parameters of the weight distribution module and the video matching module can be trained simultaneously. In this embodiment, after the first matching degree is less than a preset matching degree threshold, the parameters of the weight distribution module and the video matching module are adjusted until the adjusted matching degree is less than the preset matching degree threshold to obtain a trained video search model.

[0059] Wherein, after each adjustment of the model parameters of the initial video search model, the sample video retrieval text is input into the adjusted video search model, the second sample retrieval text feature of the sample video retrieval text is extracted by the adjusted video search model, and multiple second sample feature weights corresponding to the multimodal video information are assigned by the adjusted video search model according to the second sample retrieval text feature, that is, the weights assigned by the adjusted weight distribution module will change, and the second sample video matching the second sample retrieval text feature and the multiple second sample feature weights is determined by the adjusted video search model according to the preset video feature library, and then the second matching degree of the second sample video and the sample video retrieval text is calculated, and it is determined whether the second matching degree is less than the preset matching degree threshold. When it is not less than the preset matching degree threshold, the adjusted video search model is determined to be a trained video search model. On the contrary, when it is less than the preset matching degree threshold, the model parameters of the adjusted video search model are adjusted until the trained video search model is obtained when the adjusted matching degree is less than the preset matching degree threshold.

[0060] Therefore, in the embodiments of the present disclosure, the video retrieval model is enabled to dynamically adjust the degree of dependence on the features corresponding to each video modality information according to the specific content of the input video retrieval text.

[0061] In summary, the training and processing method of the video search model of the embodiment of the present disclosure obtains a sample video retrieval text, and inputs the sample video retrieval text into a pre-built initial video search model, extracts the first sample retrieval text feature of the sample video retrieval text through the initial video search model, and then, assigns multiple first sample feature weights corresponding to the preset multimodal video information according to the first sample retrieval text feature through the initial video search model, determines the first sample video matching the first sample retrieval text feature and multiple first sample feature weights according to the preset video feature library through the initial video search model, calculates the first matching degree between the first sample video and the sample video retrieval text, determines whether the first matching degree is less than the preset matching degree threshold, and when it is less than the preset matching degree threshold, adjusts the model parameters of the initial video search model until the adjusted matching degree is less than the preset matching degree threshold to obtain the trained video search model. In the present technical solution, the video retrieval text is adapted to dynamically determine the multiple feature weights corresponding to the multimodal video information, which ensures the matching degree between the searched video clip and the video retrieval text, and improves the video retrieval effect.

[0062] After obtaining the successfully trained video retrieval model, video retrieval processing is also performed based on the video retrieval model. Figure 2 Flow chart of a video search processing method according to an embodiment of the present disclosure. Figure 2 As shown, the method includes:

[0063] Step 201, in response to obtaining a video retrieval instruction carrying a video retrieval text, input the video retrieval text into a pre-trained video search model.

[0064] Step 202: Determine a target video that matches the video retrieval text through a video search model according to a preset video feature library.

[0065] The video search model in this embodiment is obtained according to the training processing method of the aforementioned video search model.

[0066] In one embodiment of the present disclosure, retrieval text features corresponding to video retrieval text are extracted through a video search model, multiple feature weights corresponding one-to-one to preset multimodal video information are determined based on the retrieval text features, and a matching target video is determined in a preset video feature library based on the retrieval text features and the multiple feature weights through the video search model.

[0067] In this embodiment, the feature similarity between the first sample retrieval text feature and each video feature in each video feature set in the preset video feature library can be calculated, the product value of each feature similarity and the corresponding first sample retrieval text feature can be calculated, the sum of all product values ​​corresponding to each video feature set can be calculated, and all sum values ​​corresponding to all video feature sets can be sorted in descending order to obtain a sorting result, and the sample video corresponding to the first sum value in the sorting result is determined to be the matching target video.

[0068] In this embodiment, multiple feature weights corresponding to preset multimodal video information are determined based on the retrieval text features, and the multiple feature weights are not exactly the same. Thus, the video retrieval model can dynamically adjust the degree of dependence on the features corresponding to each video modal information according to the specific content of the input video retrieval text.

[0069] In summary, the video retrieval processing method of the embodiment of the present disclosure, in response to obtaining a video retrieval instruction carrying a video retrieval text, inputs the video retrieval text into a pre-trained video search model, and then determines the target video matching the video retrieval text through the video search model according to a preset video feature library. In this technical solution, the matching degree between the searched video clip and the video retrieval text is guaranteed, and the video retrieval effect is improved.

[0070] In order to implement the above embodiment, the present disclosure also proposes a training and processing device for a video search model. Figure 3 FIG. 1 is a structural diagram of a training processing device for a video search model according to an embodiment of the present disclosure. Figure 3 As shown, the device includes: an acquisition module 310, an extraction module 320, an allocation module 330, a first determination module 340, a second determination module 350, and a training processing module 360, wherein:

[0071] An acquisition module 310 is used to acquire sample video retrieval text;

[0072] An extraction module 320, configured to input the sample video retrieval text into a pre-built initial video search model, and extract a first sample retrieval text feature of the sample video retrieval text through the initial video search model;

[0073] An allocation module 330 is used to allocate a plurality of first sample feature weights corresponding to preset multimodal video information one by one according to the first sample retrieval text feature through an initial video search model;

[0074] A first determination module 340 is used to determine a first sample video that matches a first sample search text feature and a plurality of first sample feature weights according to a preset video feature library through an initial video search model;

[0075] A second determination module 350 is used to calculate a first matching degree between the first sample video and the sample video search text, and determine whether the first matching degree is less than a preset matching degree threshold;

[0076] The training processing module 360 ​​is used to adjust the model parameters of the initial video search model when the matching degree is less than a preset matching degree threshold, until the trained video search model is obtained when the adjusted matching degree is less than the preset matching degree threshold.

[0077] The video search model training processing device provided in the embodiments of the present disclosure can execute the video search model training processing method provided in any embodiment of the present disclosure, and has the corresponding functional modules and beneficial effects of the execution method.

[0078] In order to implement the above embodiment, the present disclosure also proposes a video search processing device.

[0079] Figure 4 The structure diagram of a video search processing device provided by an embodiment of the present disclosure is shown in FIG. 1 , which can be implemented by software and / or hardware and can generally be integrated in an electronic device. Figure 4 As shown, the device includes: an output module 410 and a third determination module 420, wherein the output module 410 is used to input the video retrieval text into a pre-trained video search model in response to obtaining a video retrieval instruction carrying a video retrieval text;

[0080] The third determination module 420 is used to determine the target video matching the video retrieval text through the video search model according to a preset video feature library, wherein the video search model is obtained according to the aforementioned video search model training processing method.

[0081] The video search processing device provided in the embodiments of the present disclosure can execute the video search processing method provided in any embodiment of the present disclosure, and has the corresponding functional modules and beneficial effects of the execution method.

[0082] In order to implement the above embodiments, the present disclosure also proposes a computer program product, including a computer program / instruction, which, when executed by a processor, implements the training processing method of the video search model in the above embodiments, or the video search processing method.

[0083] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure.

[0084] The following specific reference Figure 5, which shows a schematic diagram of the structure of an electronic device 500 suitable for implementing the embodiment of the present disclosure. The electronic device 500 in the embodiment of the present disclosure may include but is not limited to mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle terminals (such as vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 5 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0085] like Figure 5 As shown, the electronic device 500 may include a processor (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a memory 508 to a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device 500 are also stored. The processor 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0086] Typically, the following devices may be connected to the I / O interface 505: an input device 506 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 507 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 508 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 509. The communication device 509 may allow the electronic device 500 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 5 The electronic device 500 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0087] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 509, or installed from the memory 508, or installed from the ROM 502. When the computer program is executed by the processor 501, the training processing method of the video search model of the embodiment of the present disclosure is executed, or the above-mentioned functions defined in the video search processing method are executed.

[0088] It should be noted that the computer-readable medium disclosed above may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0089] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0090] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0091] The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device:

[0092] The sample video retrieval text is obtained, and the sample video retrieval text is input into a pre-built initial video search model, and the first sample retrieval text feature of the sample video retrieval text is extracted through the initial video search model, and then, the initial video search model is used to assign multiple first sample feature weights corresponding to the preset multimodal video information according to the first sample retrieval text feature, and the initial video search model is used to determine the first sample video matching the first sample retrieval text feature and the multiple first sample feature weights according to the preset video feature library, and the first matching degree between the first sample video and the sample video retrieval text is calculated, and it is determined whether the first matching degree is less than the preset matching degree threshold. When it is less than the preset matching degree threshold, the model parameters of the initial video search model are adjusted until the adjusted matching degree is less than the preset matching degree threshold. The trained video search model is obtained. In this technical solution, the video retrieval text is adapted to dynamically determine the multiple feature weights corresponding to the multimodal video information, which ensures the matching degree between the searched video clip and the video retrieval text, and improves the video retrieval effect. Alternatively, in response to obtaining a video retrieval instruction carrying the video retrieval text, the video retrieval text is input into a pre-trained video search model, and then, the target video matching the video retrieval text is determined by the video search model according to the preset video feature library.

[0093] The electronic device may be written in one or more programming languages ​​or a combination thereof to write computer program code for performing the operations of the present disclosure, including but not limited to object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0094] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0095] The units involved in the embodiments described in the present disclosure may be implemented by software or hardware, wherein the name of a unit does not, in some cases, limit the unit itself.

[0096] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0097] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0098] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other to form a technical solution.

[0099] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0100] Although the subject matter has been described in language specific to structural features and / or methodological logical actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing the claims.

Claims

1. A training method for a video search model, characterized in that: The following steps are involved: Obtaining a sample video retrieval text, and inputting the sample video retrieval text into a pre-built initial video search model, and extracting a first sample retrieval text feature of the sample video retrieval text through the initial video search model; Allocating a plurality of first sample feature weights corresponding to preset multimodal video information one by one according to the first sample retrieval text feature through the initial video search model; Determine, by means of the initial video search model, a first sample video that matches the first sample search text feature and the weights of the plurality of first sample features according to a preset video feature library; Calculating a first matching degree between the first sample video and the sample video search text, and determining whether the first matching degree is less than a preset matching degree threshold; When the degree of matching is less than the preset matching threshold, the model parameters of the initial video search model are adjusted until the adjusted degree of matching is less than the preset matching threshold to obtain a trained video search model.

2. The method according to claim 1, characterized in that The adjusting the model parameters of the initial video search model until the adjusted matching degree is less than the preset matching degree threshold to obtain a trained video search model includes: After adjusting the model parameters of the initial video search model, the sample video search text is input into the adjusted video search model, and a second sample search text feature of the sample video search text is extracted through the adjusted video search model; Allocating a plurality of second sample feature weights corresponding one-to-one to the multimodal video information according to the second sample retrieval text feature through the adjusted video search model; Determining, by the adjusted video search model and according to the preset video feature library, a second sample video that matches the second sample search text feature and the plurality of second sample feature weights; Calculating a second matching degree between the second sample video and the sample video search text, and determining whether the second matching degree is less than the preset matching degree threshold; When the value is not less than the preset matching degree threshold, the adjusted video search model is determined to be the trained video search model.

3. The method according to claim 2, characterized in that After determining whether the second matching degree is less than the preset matching degree threshold, the method further includes: When the degree of match is less than the preset matching degree threshold, the model parameters of the adjusted video search model are adjusted until the trained video search model is obtained when the adjusted degree of match is less than the preset matching degree threshold.

4. The method according to claim 1, characterized in that The preset video feature library includes multiple video feature sets corresponding one-to-one to multiple sample videos, and each of the video feature sets includes multiple video features corresponding one-to-one to the corresponding sample video and the multimodal video information.

5. The method according to claim 4, characterized in that The determining, by the initial video search model according to a preset video feature library, a first sample video that matches the first sample search text feature and the weights of the plurality of first sample features comprises: Calculating feature similarity between the first sample retrieval text feature and each of the video features in each of the video feature sets; Calculating the product value of each of the feature similarities and the corresponding first sample feature weight; Calculating a sum of all the product values ​​corresponding to each of the video feature sets; The matched first sample video is determined according to the sum value.

6. The method according to claim 5, characterized in that The determining the matched first sample video according to the sum value includes: Sorting all the sum values ​​corresponding to all the video feature sets in descending order to obtain a sorting result; Determine that the sample video corresponding to the first sum value in the sorting result is the matched first sample video.

7. The method according to any one of claims 1 to 6, characterized in that: The calculating a first matching degree between the first sample video and the sample video retrieval text includes: According to each type of loss function in the preset multi-type loss function, respectively calculate the loss value of the first sample video and the sample video retrieval text to obtain a plurality of loss values ​​corresponding to the multi-type loss function; An average of the multiple loss values ​​is calculated to obtain the first matching degree.

8. The method according to any one of claims 1 to 6, characterized in that: The multimodal video information includes multiple of the following: Video picture information, audio information, video text description information, dialogue subtitle information, background music, and video evaluation information.

9. A video search processing method, characterized in that: The following steps are involved: In response to obtaining a video retrieval instruction carrying a video retrieval text, inputting the video retrieval text into a pre-trained video search model; The target video matching the video retrieval text is determined by the video search model according to a preset video feature library, wherein the video search model is obtained according to the training processing method of the video search model described in any one of claims 1-8.

10. The method according to claim 9, characterized in that The step of determining a target video matching the video search text according to a preset video feature library by using the video search model includes: Extracting retrieval text features corresponding to the video retrieval text through the video search model, and determining a plurality of feature weights corresponding to preset multimodal video information one by one according to the retrieval text features; The video search model determines the matching target video in a preset video feature library according to the retrieved text features and the multiple feature weights.

11. A training and processing device for a video search model, characterized in that: include: An acquisition module is used to acquire sample video retrieval text; An extraction module, used for inputting the sample video retrieval text into a pre-built initial video search model, and extracting a first sample retrieval text feature of the sample video retrieval text through the initial video search model; An allocation module, configured to allocate a plurality of first sample feature weights corresponding to preset multimodal video information one by one according to the first sample retrieval text feature through the initial video search model; A first determination module is used to determine a first sample video that matches the first sample search text feature and the weights of the plurality of first sample features according to a preset video feature library through the initial video search model; A second determination module is used to calculate a first matching degree between the first sample video and the sample video search text, and determine whether the first matching degree is less than a preset matching degree threshold; The training processing module is used to adjust the model parameters of the initial video search model when the matching degree is less than the preset matching degree threshold, until the trained video search model is obtained when the adjusted matching degree is less than the preset matching degree threshold.

12. A video search processing device, characterized in that: include: An output module, configured to, in response to obtaining a video retrieval instruction carrying a video retrieval text, input the video retrieval text into a pre-trained video search model; A third determination module is used to determine a target video that matches the video retrieval text through the video search model according to a preset video feature library, wherein the video search model is obtained according to the training processing method of the video search model described in any one of claims 1-8.

13. An electronic device, characterized in that: The electronic device comprises: processor; a memory for storing instructions executable by the processor; The processor is used to read the executable instructions from the memory and execute the executable instructions to implement the training processing method of the video search model described in any one of claims 1-8 above, or the video search processing method described in claim 9 or 10.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is used to execute the training processing method of the video search model described in any one of claims 1-8 above, or the video search processing method described in claim 9 or 10.

Images