Method, apparatus, device and storage medium for information processing

Abstract

According to embodiments of the present disclosure, a method, an apparatus, a device and a storage medium for information processing are provided. The method comprises: obtaining a plurality of pieces of response content generated by a first model based on input information; constructing, by using a second model, a plurality of groups of questions corresponding to the plurality of pieces of response content respectively; generating an evaluation result for the plurality of pieces of response content based on answer information of the plurality of groups of questions by the first model; and training the first model by using at least one pair of response content in the plurality of pieces of response content based on the evaluation result. Embodiments of the present disclosure can construct questions corresponding to the response content, and determine the evaluation result of the response content according to the response content generated by the first model for the first time and the answer information generated for the second time, so as to train the model, thereby improving the accuracy of the output content of the model.

Description

Technical Field

[0001] The exemplary embodiments disclosed herein generally relate to the field of computers, and particularly to methods, apparatus, devices and computer-readable storage media for information processing. Background Technology

[0002] With the development of computer technology, network models are able to provide more and more functions. For example, visual language models are deep learning models that can process visual and linguistic information simultaneously and achieve cross-modal understanding, reasoning, or generation. By jointly learning the association between vision and language, they can complete multimodal tasks such as image description generation, visual question answering, and image-text retrieval. However, the accuracy of the output content in current visual language models is still insufficient. Summary of the Invention

[0003] In a first aspect of this disclosure, an information processing method is provided. The method includes: acquiring multiple response contents generated by a first model based on input information; constructing multiple sets of questions corresponding to the multiple response contents using a second model; generating evaluation results for the multiple response contents based on the first model's responses to the multiple sets of questions; and training the first model based on the evaluation results using at least one pair of response contents from the multiple response contents.

[0004] In a second aspect of this disclosure, an information processing apparatus is provided. The apparatus includes: an acquisition module configured to acquire multiple response contents generated by a first model based on input information; a construction module configured to construct multiple sets of questions corresponding to the multiple response contents using a second model; a generation module configured to generate evaluation results for the multiple response contents based on the first model's responses to the multiple sets of questions; and a training module configured to train the first model based on the evaluation results using at least one pair of response contents from the multiple response contents.

[0005] In a third aspect of this disclosure, an electronic device is provided. The device includes at least one processing unit; and at least one memory coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit. When executed by the at least one processing unit, the instructions cause the device to perform the method of the first aspect.

[0006] In a fourth aspect of this disclosure, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program that can be executed by a processor to implement the method of the first aspect.

[0007] In a fifth aspect of this disclosure, a computer program product is provided. The computer program product includes computer-executable instructions that, when executed by a processor, implement the method according to a first aspect of this disclosure.

[0008] It should be understood that the content described in this content section is not intended to limit the key or essential features of the embodiments of this disclosure, nor is it intended to restrict the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0009] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0010] Figure 1 A schematic diagram of an example environment in which embodiments of the present disclosure can be implemented is shown;

[0011] Figure 2 A flowchart illustrating an information processing procedure according to some embodiments of the present disclosure is shown;

[0012] Figure 3 A flowchart describing a task according to some embodiments of the present disclosure is shown;

[0013] Figure 4 A flowchart of a question-and-answer task according to some embodiments of this disclosure is shown;

[0014] Figure 5 A flowchart illustrating preference optimization according to some embodiments of this disclosure is shown;

[0015] Figure 6 A schematic structural block diagram of an information processing apparatus according to certain embodiments of the present disclosure is shown;

[0016] Figure 7 A block diagram of an electronic device capable of implementing several embodiments of the present disclosure is shown. Detailed Implementation

[0017] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0018] It should be noted that the headings of any section / subsection provided herein are not limiting. Various embodiments are described throughout this document, and embodiments of any type may be included under any section / subsection. Furthermore, embodiments described in any section / subsection may be combined in any way with any other embodiments described in the same section / subsection and / or different sections / subsections.

[0019] In the description of embodiments of this disclosure, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The term "some embodiments" should be understood as "at least some embodiments". Other explicit and implicit definitions may also be included below. The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0020] The embodiments of this disclosure may involve user data, data acquisition, and / or use. All of these aspects comply with applicable laws, regulations, and relevant provisions. In the embodiments of this disclosure, all data collection, acquisition, processing, manipulation, forwarding, and use are conducted with the user's knowledge and confirmation. Accordingly, in implementing the embodiments of this disclosure, the type, scope of use, and usage scenarios of any data or information that may be involved should be communicated to the user and their authorization obtained in accordance with relevant laws and regulations through appropriate means. The specific methods of notification and / or authorization may vary depending on the actual situation and application scenario, and the scope of this disclosure is not limited in this respect.

[0021] In this specification and the embodiments, any processing of personal information will be carried out only under the premise of legality (such as obtaining the consent of the personal information subject, or being necessary for the performance of a contract), and will only be carried out within the scope stipulated or agreed upon. A user's refusal to process personal information other than that necessary for basic functions will not affect the user's use of basic functions.

[0022] Visual language models are artificial intelligence models that combine computer vision and natural language processing. They can simultaneously understand and process visual and textual information and establish connections between the two. However, during content generation, visual language models often generate content that is inconsistent with, does not exist, or is false. This situation is called model hallucination, and the occurrence of hallucination can greatly affect the accuracy of the model's output.

[0023] The embodiments of this disclosure propose an information processing scheme. According to the scheme, multiple response contents generated by a first model based on input information are obtained; multiple sets of questions corresponding to the multiple response contents are constructed using a second model; evaluation results for the multiple response contents are generated based on the first model's answer information to the multiple sets of questions; and the first model is trained based on the evaluation results using at least one pair of response contents from the multiple response contents.

[0024] Based on this approach, since the model's output of short answers is more accurate, embodiments of this disclosure can construct questions corresponding to the response content, have the first model answer these questions, and then determine the evaluation result of the response content based on the first response content generated by the first model and the second answer information generated by the first model to train the model. This allows the model to learn response content with better evaluation results and improve the accuracy of the model's output content.

[0025] Example Environment

[0026] Figure 1 A schematic diagram of an example environment 100 in which embodiments of the present disclosure can be implemented is shown. For example... Figure 1 As shown, the example environment 100 may include an electronic device 110. In some embodiments, the electronic device 110 may acquire multiple response contents generated by a first model based on input information; construct multiple sets of questions corresponding to the multiple response contents using a second model; generate evaluation results for the multiple response contents based on the first model's answer information to the multiple sets of questions; and train the first model using at least one pair of response contents from the multiple response contents based on the evaluation results. Model 120 may be deployed on the electronic device 110 or on other devices, which will not be elaborated here.

[0027] In some embodiments, the electronic device 110 may be any type of mobile terminal, fixed terminal, or portable terminal, including mobile phones, desktop computers, laptop computers, notebook computers, netbook computers, tablet computers, media computers, multimedia tablets, handheld computers, portable gaming terminals, VR / AR devices, personal communication system (PCS) devices, personal navigation devices, personal digital assistants (PDAs), audio / video players, digital cameras / camcorders, positioning devices, television receivers, radio receivers, e-book devices, gaming devices, or any combination thereof, including accessories and peripherals of these devices or any combination thereof. In some embodiments, the electronic device 110 may also support any type of interface for the target user (such as "wearable" circuitry).

[0028] Electronic device 110 can also be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, and big data and artificial intelligence platforms. Electronic device 110 may include, for example, computing systems / servers, such as mainframes, edge computing nodes, computing devices in cloud environments, and so on.

[0029] It should be understood that the structure and function of the various embedded elements in environment 100 are described for illustrative purposes only and do not imply any limitation on the scope of this disclosure.

[0030] The following description will continue with reference to the accompanying drawings, which will provide some exemplary embodiments of this disclosure.

[0031] Example process

[0032] Figure 2 A flowchart of an information processing procedure 200 according to some embodiments of the present disclosure is shown. Procedure 200 can be implemented at electronic device 110. Reference is made below. Figure 1 Describe the process 200.

[0033] In box 210, electronic device 110 can acquire multiple response contents generated by the first model based on the input information.

[0034] In some embodiments, the first model can be a Vision-Language Model (VLM), an artificial intelligence model that combines computer vision and natural language processing. It can simultaneously understand and process visual and textual information and establish the relationship between the two. VLMs can handle various tasks, such as description tasks, question-answering tasks, and so on.

[0035] Because visual language models often generate content that is inconsistent with, does not exist, or is false during the content generation process, this phenomenon is called model hallucination, and the occurrence of hallucination can significantly affect the accuracy of the model's output. For visual language models, hallucination is more likely to occur when generating lengthy descriptive responses, while it is less likely to occur when generating short, direct responses. In other words, the short, direct responses output by the model are more accurate than the lengthy descriptive responses. Therefore, we can use the model to generate both short, direct responses and lengthy descriptive responses, and then compare these two types of responses using the model's own evaluation system to optimize the model itself.

[0036] Due to the randomness of the content generated by the model, the input information can be processed multiple times using the first model to obtain different multiple generation results, which can then be used as multiple response contents.

[0037] In some embodiments, the input information includes media content and prompts associated with the media content. For example, an image of a dining table could be used as the media content, and the prompt could be "Please describe this image in detail."

[0038] In some embodiments, when the first model processes the description task, the prompt message can be configured to instruct the first model to generate descriptive text for media content. For example, an image of a dining table can be used as media content, and the prompt message could be "Please describe this image in detail."

[0039] In some embodiments, the electronic device 110 may provide media content and prompt information to the first model to obtain multiple response contents generated by the first model, the multiple response contents including multiple media description texts for the media content. For example, refer to Figure 3 The system can provide the first model 303 with an image of a dining table as media content 301, along with a prompt message 302: "Please describe this image in detail." The response content 304 generated by the first model can be: "The image shows a long dining table in the center, surrounded by many black chairs. The brick wall adds a rustic feel to the space. In addition to the dining area, the room also includes a kitchen area with a sink, oven, and refrigerator. The kitchen contains various utensils, such as spoons, knives, bowls, etc." This media description text for the media content can be called the first response content.

[0040] In some embodiments, in order to collect more response content to build preference pairs for training the model in subsequent steps, the first model can generate multiple response contents y1, y2, ..., y3 for the same media content v and cue information x. N y~p θ The responses (·|x,v) will not be entirely identical due to factors such as model parameters. These responses are lengthy descriptive responses, and therefore may contain accurate responses or misleading ones.

[0041] In some embodiments, when the first model processes the question-answering task, the prompt information can also be query text associated with media content. For example, an image of a bicycle with a basket can be used as media content, and the query text could be "What's in the basket?".

[0042] In some embodiments, the electronic device 110 may provide media content v and query text q to a first model to obtain media description text c for the media content, the media description text being associated with the query text. Then, it obtains response text y, y ~ p, generated by the first model based on the media content and media description text, for the query text. θ (·|q,c,v) is used as the second response content {c,y} in response to the input information.

[0043] The response text, generated by the first model based on media content and media description text, is the answer to the query text. Because visual language models typically produce brief responses in such question-and-answer tasks, it's difficult to extract sufficient atomic facts from these short responses for subsequent self-consistent response evaluation. Therefore, to collect more detailed responses and richer factual information, media description text is also needed. This media description text, associated with the query text, is more detailed and richer in content, helping to extract more information in subsequent steps. The media description text and the response text together constitute the second response content.

[0044] For example, refer to Figure 4 An image of a bicycle with a basket can be used as media content 401. The query text 402 can be "What's in the basket?". Inputting the image and query text into the first model 403 will produce media description text 404 and response text 405. Media description text 404 can be "A big watermelon... Next to it, there is a plant with purple and white flowers, it seems to be a potted plant with a dark green or black flowerpot. There are also some baskets filled with something that looks like hay or straw." Response text 405 can be "The picture shows a bicycle with a basket containing a watermelon, a potted plant with purple and white flowers, and baskets of hay or straw."

[0045] In box 220, electronic device 110 can use the second model to construct multiple sets of questions corresponding to multiple response contents.

[0046] In one embodiment, the second model is used to identify objects or semantic information in the text, where the second model can be a grammatical rule or a language model. For example, if the text description is "There is a table in the image," then the second model can determine that the text description includes the object "table." Or, if the text description is "There are white flowers in the basket," then the second model can determine that the flowers in the text description are white.

[0047] In some embodiments, a second model can be used to construct multiple questions, which are then answered by a first model to obtain a brief, direct response generated by the first model. While the model processes the descriptive task, the electronic device 110 can provide the second model with a first response from among multiple response contents to identify multiple objects associated with the media content. Based on these multiple objects, a first set of questions corresponding to the first response content is then constructed, configured to query whether the media content includes the corresponding object.

[0048] For example, refer to Figure 3 If media content 301 is an image of a dining table, the prompt message 302 is "Please describe this image in detail." The response content 304 generated in the first model 303 is "The image shows a long dining table in the center, surrounded by many black chairs. The brick wall adds a rustic feel to the space. In addition to the dining area, the room also includes a kitchen area with a sink, oven, and refrigerator. The kitchen contains various utensils such as spoons, knives, bowls...". Then, the second model can be used to extract the objects included in the response content: "dining table, chairs, brick wall, dining area, kitchen area, sink, oven, refrigerator, spoon, knife, bowl." Then, a predefined template can be used to construct the first set of questions 305 based on these objects, such as "Does the image include a dining table?" or "Does the image include a sink?" etc.

[0049] In some embodiments, when the model processes a question-and-answer task, a two-round thought chain prompting method can be used. The electronic device 110 can provide the second model with media description text and response text to construct a second set of questions corresponding to the second response content. That is, the second set of questions is related to both the media description text and the response text.

[0050] In some embodiments, the second model can be configured to: extract semantic information associated with the media content based on the media description text and the response text, and then construct a second set of questions associated with the semantic information. The second set of questions is used to inquire whether the description content corresponding to the semantic information is accurate.

[0051] In one embodiment, a second set of questions can be constructed by extracting key information from the media description text and the response text. Key information can include various types: content composition, objects, the number of objects, actions, relationships between objects, object attributes, image content inference, etc. For example, content composition can include knowledge and functions. Objects can be nouns, such as people, animals, plants, etc. Relationships between objects can include direct contact, interactive behaviors, positional relationships, etc. Actions can include running, walking, etc. Object attributes can include size, shape, color, material, etc. Image content inference can include the number of objects, object relationships, object actions, etc., conveyed by the image. Key information can also include text in the image, clock readings, etc., without limitation.

[0052] In some embodiments, questions can be constructed as binary questions, that is, questions that can be answered with "yes" or "no". If the constructed questions contain complex objects or relationships, such questions can be broken down into multiple binary questions. The model answers these simple binary questions more accurately, and binary questions also make the question content clearer, thus avoiding question duplication. Furthermore, multiple questions can be posed for a single object, as long as they are not repeated.

[0053] For example, refer to Figure 4 Media content 401 is an image of a bicycle, query text 402 is "What's in the basket?", media description text 404 is "A big watermelon...Next to it, there's a plant with purple and white flowers, it seems to be a potted plant with a dark green or black flowerpot. There are also some baskets filled with something that looks like hay or straw.", and response text 405 is "The picture shows a bicycle with a basket containing a watermelon, a potted plant with purple and white flowers, and baskets of hay or straw." The second model can then extract semantic information associated with the media content based on the media description text and response text, and construct a second set of questions 406, such as "Are the flowers purple and white?", "Is the watermelon in the basket?", etc.

[0054] In box 230, electronic device 110 can generate evaluation results for multiple response contents based on the first model's answer information for multiple sets of questions.

[0055] In one embodiment, during task processing, a situation may arise where the first model generates a response stating "There is a dining table in the image." The second model, based on this response, generates a question asking "Does the image include a dining table?" The first model answers "No" based on the image. This results in inconsistencies between the first model's two responses. Furthermore, since shorter responses are more accurate, it's possible that the first generated response contains a misinterpretation. Therefore, the self-evaluation ability of the first model can be utilized to optimize the model.

[0056] In some embodiments, the electronic device 110 can generate an evaluation result for multiple response contents based on the comparison of the answer information and reference answers corresponding to multiple sets of questions. The reference answer is the answer obtained based on the response content. For example, the response content generated in the first model is "There is a long dining table in the middle of the image...", and the first set of questions constructed using the second model includes "Does the image include a dining table?", the reference answer is "yes", because it can be known from the response content that the image includes a dining table. Then, the answer information of the first model to the question "Does the image include a dining table?" can be obtained. If the answer information is "yes", it indicates that there is internal consistency between the response content and the answer information, and it can be determined that the response content is accurate at this time; if the answer information is "no", it indicates that there is internal inconsistency between the response content and the answer information, and it can be determined that the response content is a hallucination at this time.

[0057] Next, the response content is evaluated based on its internal consistency with the answer information, resulting in evaluation results for multiple responses. If any internal inconsistencies exist between the response content and the answer information, they can be recorded using a negative response "No," for reference. Figure 3 Evaluation results 306 and Figure 4 The evaluation result is 407. The more negative responses there are in the evaluation result, the greater the possibility that the response content contains hallucinations, and the lower the accuracy of the response content; the fewer negative responses there are in the evaluation result, the less likely that the response content contains hallucinations, and the higher the accuracy of the response content.

[0058] In box 240, electronic device 110 can train a first model based on the evaluation results, using at least one pair of response contents from multiple response contents.

[0059] In some embodiments, the model can be trained using preference optimization methods to improve the accuracy of the model's output. Direct Preference Optimization (DPO) is a method for aligning language models with human preferences. By directly optimizing model parameters, it makes the probability distribution of the output more consistent with human-annotated preference data, thereby simplifying the training process and improving stability.

[0060] In some embodiments, the electronic device 110 may determine at least one pair of response contents from a plurality of response contents based on evaluation results, wherein each pair of response contents includes two response contents corresponding to different evaluations. For example, refer to Figure 5 Based on multiple evaluation results 501, at least one pair of response content can be determined from multiple response content. Each pair of response content 502 is a pair of preference pairs. The preference pair includes preferred response content y+ and unpreferred response content y-. The evaluation results corresponding to the preferred response content have fewer negative response content, while the evaluation results corresponding to the unpreferred response content have more negative response content.

[0061] In some embodiments, the electronic device 110 can determine positive and negative response content based on the evaluation results of at least one pair of response content, where the evaluation of positive response content is superior to that of negative response content. Then, a first model is trained using the positive and negative response content. For example, referring to... Figure 5 The first model 503 can generate three response items based on the input information 504. Then, the evaluation results corresponding to these three response items are obtained, and the three response items are ranked according to the frequency of negative response items in the evaluation results. At least one pair of response items is obtained as a preference pair based on the ranking results. The objective function for preference optimization is as follows:

[0062]

[0063] X represents the prompt message, v represents the media content, y+ represents the preferred response content, y- represents the non-preferred response content, and p represents the p message. ref Let θ represent the pre-trained visual language model, β be the hyperparameter scaling factor, and σ be the sigmoid function. This optimization process makes the model's output distribution tend towards an internally consistent and better output.

[0064] Because the model is less likely to produce illusions and has higher accuracy when generating short responses, embodiments of this disclosure can first use a first model to generate response content, then use a second model to construct questions based on the response content, and have the first model answer these questions. This allows for evaluation of the response content based on the internal consistency between the model's first and second output responses. In this way, the model can learn whether there are illusions in the previously generated response content without external supervision, through a self-reflection mechanism, thus effectively using unlabeled data. Then, based on the evaluation results, multiple preference pairs are determined to train the model, enabling it to learn how to output response content with fewer illusions and higher accuracy, thereby improving the accuracy of the model's output content.

[0065] Example devices and equipment

[0066] Embodiments of this disclosure also provide corresponding apparatus for implementing the above methods or processes. Figure 6 A schematic structural block diagram of a device 600 for touch interaction according to certain embodiments of the present disclosure is shown. The device 600 may be implemented as or included in the electronic device 110 discussed above. The various modules / components in the device 600 may be implemented by hardware, software, firmware, or any combination thereof.

[0067] like Figure 6 As shown, the device 600 includes an acquisition module 610 configured to acquire multiple response contents generated by a first model based on input information; a construction module 620 configured to construct multiple sets of questions corresponding to the multiple response contents using a second model; a generation module 630 configured to generate evaluation results for the multiple response contents based on the first model's answer information to the multiple sets of questions; and a training module 640 configured to train the first model based on the evaluation results using at least one pair of response contents from the multiple response contents.

[0068] In some embodiments, the input information includes media content and prompts associated with the media content.

[0069] In some embodiments, the acquisition module 610 is configured to provide media content and prompt information to the first model to acquire multiple response contents generated by the first model, the multiple response contents including multiple media description texts for the media content.

[0070] In some embodiments, the construction module 620 is configured to provide a first response content from a plurality of response contents to a second model to identify a plurality of objects associated with the media content; and based on the plurality of objects, to construct a first set of questions corresponding to the first response content, the first set of questions being configured to query whether the media content includes the corresponding object.

[0071] In some embodiments, the acquisition module 610 is configured to provide media content and query text to the first model to acquire media description text for the media content, the media description text being associated with the query text; and to acquire response text for the query text generated by the first model based on the media content and media description text, as a second response content for the input information.

[0072] In some embodiments, the construction module 620 is configured to provide media description text and response text to the second model to construct a second set of questions corresponding to the second response content.

[0073] In some embodiments, the second model is configured to: extract semantic information associated with the media content based on the media description text and the response text; and construct a second set of questions associated with the semantic information.

[0074] In some embodiments, the second set of questions is used to inquire whether the description corresponding to the semantic information is accurate.

[0075] In some embodiments, the generation module 630 is configured to generate evaluation results for multiple response contents based on the answer information and a comparison with reference answers corresponding to multiple sets of questions.

[0076] In some embodiments, the training module 640 is configured to determine at least one pair of response contents from a plurality of response contents based on the evaluation results, wherein each pair of response contents includes two response contents corresponding to different evaluations.

[0077] In some embodiments, the training module 640 is configured to determine positive response content and negative response content based on the evaluation of at least one pair of response content, wherein the evaluation of the positive response content is superior to the evaluation of the negative response content; and to train a first model using the positive response content and the negative response content.

[0078] The units included in device 600 can be implemented in various ways, including software, hardware, firmware, or any combination thereof. In some embodiments, one or more units may be implemented using software and / or firmware, such as machine-executable instructions stored on a storage medium. In addition to or as an alternative to machine-executable instructions, some or all of the units in device 600 may be implemented at least partially by one or more hardware logic components. By way of example and not 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), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0079] Figure 7 A block diagram of an electronic device 700 in which one or more embodiments of the present disclosure may be implemented is shown. It should be understood that... Figure 7 The electronic device 700 shown is merely exemplary and should not be construed as limiting the functionality and scope of the embodiments described herein. Figure 7 The electronic device 700 shown can be used to achieve Figure 1 The electronic device 110 shown.

[0080] like Figure 7As shown, electronic device 700 is in the form of a general-purpose electronic device. Components of electronic device 700 may include, but are not limited to, one or more processors or processing units 710, memory 720, storage device 730, one or more communication units 740, one or more input devices 750, and one or more output devices 760. Processing unit 710 may be a physical or virtual processor and is capable of performing various processes according to programs stored in memory 720. In a multiprocessor system, multiple processing units execute computer-executable instructions in parallel to improve the parallel processing capability of electronic device 700.

[0081] Electronic device 700 typically includes multiple computer storage media. Such media can be any accessible media that is accessible to electronic device 700, including but not limited to volatile and non-volatile media, removable and non-removable media. Memory 720 can be volatile memory (e.g., registers, cache, random access memory (RAM)), non-volatile memory (e.g., read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory), or some combination thereof. Storage device 730 can be a removable or non-removable medium and can include machine-readable media, such as flash drives, disks, or any other media that can be used to store information and / or data (e.g., training data for training) and can be accessed within electronic device 700.

[0082] Electronic device 700 may further include additional removable / non-removable, volatile / non-volatile storage media. Although not explicitly stated... Figure 7 As shown, disk drives for reading from or writing to removable, non-volatile disks (e.g., "floppy disks") and optical disk drives for reading from or writing to removable, non-volatile optical disks can be provided. In these cases, each drive can be connected to a bus (not shown) via one or more data media interfaces. Memory 720 may include computer program product 725 having one or more program modules configured to perform various methods or actions of various embodiments of this disclosure.

[0083] The communication unit 740 enables communication with other electronic devices via a communication medium. Additionally, the functionality of the components of the electronic device 700 can be implemented using a single computing cluster or multiple computing machines capable of communicating via communication connections. Therefore, the electronic device 700 can operate in a networked environment using logical connections to one or more other servers, network personal computers (PCs), or another network node.

[0084] Input device 750 can be one or more input devices, such as a mouse, keyboard, trackball, etc. Output device 760 can be one or more output devices, such as a monitor, speaker, printer, etc. Electronic device 700 can also communicate with one or more external devices (not shown) via communication unit 740 as needed. These external devices include storage devices, display devices, etc., and can communicate with one or more devices that enable user interaction with electronic device 700, or with any device that enables electronic device 700 to communicate with one or more other electronic devices (e.g., network card, modem, etc.). Such communication can be performed via input / output (I / O) interface (not shown).

[0085] According to an exemplary implementation of this disclosure, a computer-readable storage medium is provided that stores computer-executable instructions thereon, wherein the computer-executable instructions are executed by a processor to implement the methods described above. According to an exemplary implementation of this disclosure, a computer program product is also provided, which is tangibly stored on a non-transitory computer-readable medium and includes computer-executable instructions, which are executed by a processor to implement the methods described above.

[0086] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, devices, and computer program products implemented according to this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0087] These computer-readable program instructions can be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processing unit of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0088] Computer-readable program instructions can be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions that execute on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0089] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction, which contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0090] Various implementations of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed implementations. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described implementations. The terminology used herein is chosen to best explain the principles, practical applications, or improvements to technology in the market, or to enable others skilled in the art to understand the various implementations disclosed herein.

Claims

1. An information processing method, comprising: Obtain the multiple response contents generated by the first model based on the input information; Using the second model, construct multiple sets of questions corresponding to the various response contents; Based on the first model's response information to the multiple sets of questions, an evaluation result is generated for the multiple response contents; as well as Based on the evaluation results, the first model is trained using at least one pair of response contents from the plurality of response contents.

2. The method according to claim 1, wherein the input information includes media content and prompts associated with the media content.

3. The method according to claim 2, wherein the prompting information is configured to instruct the first model to generate descriptive text of the media content, and obtaining multiple response contents generated by the first model based on the input information includes: The media content and the prompt information are provided to the first model to obtain the multiple response contents generated by the first model, wherein the multiple response contents include multiple media description texts for the media content.

4. The method according to claim 3, wherein constructing multiple sets of questions corresponding to the multiple response contents using the second model includes: The first response content from the multiple response contents is provided to the second model to identify multiple objects associated with the media content; as well as Based on the multiple objects, a first set of questions corresponding to the first response content is constructed. The first set of questions is configured to query whether the media content includes the corresponding object.

5. The method of claim 2, wherein the prompting information includes query text associated with the media content, and obtaining multiple response contents generated by the first model based on the input information includes: The media content and the query text are provided to the first model to obtain a media description text for the media content, which is associated with the query text; as well as Obtain the response text generated by the first model based on the media content and the media description text in response to the query text, as the second response content for the input information.

6. The method according to claim 5, wherein constructing multiple sets of questions corresponding to the multiple response contents using the second model includes: The media description text and the response text are provided to the second model to construct a second set of questions corresponding to the second response content.

7. The method of claim 6, wherein the second model is configured as follows: Based on the media description text and the response text, semantic information associated with the media content is extracted; and Construct the second set of questions associated with the semantic information.

8. The method according to claim 7, wherein the second set of questions is used to inquire whether the description content corresponding to the semantic information is accurate.

9. The method according to claim 1, wherein generating an evaluation result for the multiple response contents based on the first model's answer information to the multiple sets of questions includes: Based on the comparison between the answer information and the reference answers corresponding to the multiple sets of questions, the evaluation results for the multiple response contents are generated.

10. The method according to claim 1, further comprising: Based on the evaluation results, at least one pair of response contents is determined from the plurality of response contents, wherein each pair of response contents includes two response contents corresponding to different evaluations.

11. The method of claim 1, wherein training the first model using at least one pair of response contents from the plurality of response contents comprises: Based on the evaluation of the at least one pair of response contents, positive response content and negative response content are determined, wherein the evaluation of the positive response content is superior to the evaluation of the negative response content; as well as The first model is trained using the positive response content and the negative response content.

12. An information processing apparatus, comprising: The acquisition module is configured to acquire multiple response contents generated by the first model based on the input information; The construction module is configured to use the second model to construct multiple sets of questions corresponding to the multiple response contents, respectively. The generation module is configured to generate evaluation results for the multiple response contents based on the first model's answer information to the multiple sets of questions; as well as The training module is configured to train the first model based on the evaluation results, using at least one pair of response contents from the plurality of response contents.

13. An electronic device, comprising: At least one processing unit; as well as At least one memory coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions causing the electronic device to perform the method according to any one of claims 1 to 11 when executed by the at least one processing unit.

14. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method according to any one of claims 1 to 11.

15. A computer program product comprising computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the method according to any one of claims 1 to 11.

Images