Intelligent question-answering method, device, electronic device, and storage medium based on large models

Abstract

The present disclosure proposes a large-scale model-based intelligent question-answering method, device, electronic device, and storage medium, which relate to the field of artificial intelligence technology, specifically the field of large-scale models and intelligent question-answering, including: obtaining historical question-answering information, and generating multiple candidate answer information based on the historical question-answering information through a generative large-scale model; predicting the conversion rate of the multiple candidate answer information to obtain the conversion rate information of the candidate answer information; and determining the target answer information for output based on the conversion rate information of the candidate answer information. Therefore, the present disclosure determines the target answer information for output based on the conversion rate information of the candidate answer information, thereby ensuring the accuracy of the target answer information, improving the experience during the question-answering process, and improving the conversion rate brought by the target answer information in the intelligent question-answering scenario.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, specifically to technical fields such as large models and intelligent question answering, and in particular to an intelligent question answering method, device, electronic device and storage medium based on a large model. Background Art

[0002] In related technologies, multiple answers are often generated based on contextual question and answer information based on a generative large model, and the multiple answers are scored based on a feedback model. The final answer is determined from the multiple answers based on the scoring results. However, the above method does not take into account the conversion rate of the final answer, resulting in the conversion rate of the final answer being often low. Therefore, how to improve the conversion rate of the output answer while ensuring the accuracy of the output answer has become an urgent problem to be solved. Summary of the Invention

[0003] The present disclosure proposes a large-model-based intelligent question-answering method, device, electronic device, storage medium, and computer program product.

[0004] According to the first aspect of the present disclosure, a large-model-based intelligent question-answering method is proposed, including: obtaining historical question-answering information, and generating multiple candidate reply information based on the historical question-answering information through a generative large-model; predicting the conversion rates of the multiple candidate reply information, and obtaining the conversion rate information of the candidate reply information; and determining the target reply information for output based on the conversion rate information of the candidate reply information.

[0005] According to the second aspect of the present disclosure, an intelligent question-answering device based on a large model is proposed, including: an acquisition module for acquiring historical question-answering information, and generating multiple candidate reply information based on the historical question-answering information through a generative large model; a prediction module for predicting the conversion rate of the multiple candidate reply information, and acquiring the conversion rate information of the candidate reply information; an output module for determining the target reply information for output based on the conversion rate information of the candidate reply information.

[0006] According to a third aspect of the present disclosure, an electronic device is proposed, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the large model-based intelligent question-answering method proposed in the first aspect above.

[0007] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is proposed, wherein the computer instructions are used to enable the computer to execute the large model-based intelligent question-answering method proposed in the first aspect above.

[0008] According to a fifth aspect of the present disclosure, a computer program product is proposed, comprising a computer program, which, when executed by a processor, implements the large model-based intelligent question-answering method proposed in the first aspect.

[0009] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The accompanying drawings are used to better understand the present invention and do not constitute a limitation of the present invention.

[0011] Figure 1 Schematic diagram of a flow chart of an intelligent question-answering method based on a large model according to an embodiment of the present disclosure;

[0012] Figure 2 Schematic diagram of a flow chart of an intelligent question-answering method based on a large model according to an embodiment of the present disclosure;

[0013] Figure 3 Schematic diagram of a flow chart of an intelligent question-answering method based on a large model according to an embodiment of the present disclosure;

[0014] Figure 4 This is a schematic diagram of obtaining conversion rate information of candidate answer information according to an embodiment of the present disclosure;

[0015] Figure 5 Schematic diagram of a flow chart of an intelligent question-answering method based on a large model according to an embodiment of the present disclosure;

[0016] Figure 6 Schematic diagram of the structure of an intelligent question-answering device based on a large model according to an embodiment of the present disclosure;

[0017] Figure 7 A schematic block diagram of an electronic device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0018] The following description of exemplary embodiments of the present disclosure is made in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding. These details should be considered as merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0019] Artificial Intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence.

[0020] A large model refers to a machine learning model with a large parameter scale and complexity. It requires a large amount of computing resources and storage space to train and store, and often requires distributed computing and special hardware acceleration technology. Large models have stronger generalization and expression capabilities.

[0021] Intelligent question-answering (AI) refers to an automated question-answering system implemented using artificial intelligence (AI). By analyzing, understanding, and mining large amounts of text data, it enables computers to understand and answer user questions. These systems typically incorporate key technologies such as natural language processing, knowledge representation, reasoning, and answer generation.

[0022] Figure 1 FIG. 1 is a flow chart of an intelligent question-answering method based on a large model according to an embodiment of the present disclosure. Figure 1 As shown, the method includes:

[0023] S101, obtain historical question and answer information, and generate multiple candidate answer information based on the historical question and answer information through a generative large model.

[0024] It should be noted that the execution entity of the large-scale model-based intelligent question-answering method of the present disclosure can be a hardware device capable of large-scale model-based intelligent question-answering and / or the necessary software to drive the operation of the hardware device. Optionally, the execution entity may include a workstation, server, computer, user terminal, and other intelligent device. User terminals include but are not limited to mobile phones, computers, intelligent voice interaction devices, smart appliances, and in-vehicle terminals.

[0025] Among them, historical question and answer information includes the inquiry content and answer content corresponding to different question and answer roles.

[0026] For example, in a business Q&A consultation scenario, the response content of the merchant’s customer service and the inquiry content of the target object can be obtained as historical Q&A information.

[0027] In an embodiment of the present disclosure, after obtaining historical question and answer information, a plurality of candidate reply information can be generated based on the historical question and answer information through a pre-trained generative large model.

[0028] S102: predict the conversion rates of multiple candidate answer information and obtain conversion rate information of the candidate answer information.

[0029] The conversion rate information includes a first conversion probability and a second conversion probability. The first conversion probability is used to represent the probability that the target object continues to chat, and the second conversion probability is used to represent the probability that the target object retains contact information.

[0030] In an embodiment of the present disclosure, after obtaining multiple candidate reply information, discrete feature information and text feature information corresponding to the candidate reply information can be obtained. Based on the discrete feature information and text feature information, the conversion rate of the candidate reply information can be predicted to obtain the conversion rate information of the candidate reply information.

[0031] It should be noted that when traditional deep neural networks (DNNs) are used only to predict the conversion rate of candidate answer information using discrete features, the text feature information of historical question and answer information with a large amount of information is lost. When the conversion rate of candidate answer information is predicted using only text features through a generative model, the discrete features on the target object side and the question and answer side are lost. The above method results in a low accuracy of the conversion rate information. In the embodiment of the present disclosure, discrete feature information and text feature information are used simultaneously to predict the conversion rate of candidate answer information, and the conversion rate information of the candidate answer information is obtained, thereby improving the accuracy of the conversion rate information.

[0032] S103: Determine target reply information for output based on the conversion rate information of the candidate reply information.

[0033] It should be noted that the present disclosure does not limit the specific method for determining the target response information to be output based on the conversion rate information of the candidate response information, and the method can be selected according to the actual situation.

[0034] Optionally, quality evaluation information of the candidate reply information may be obtained, and target reply information may be obtained from a plurality of candidate reply information based on the quality evaluation information and conversion rate information.

[0035] Optionally, the first conversion probability and the second conversion probability of the candidate reply information can be obtained, and the candidate reply information can be sorted using at least one of the first conversion probability and the second conversion probability as sorting reference information, and the candidate reply information ranked first can be selected and output as the target reply information.

[0036] The intelligent question-answering method based on a large model proposed in the present disclosure obtains historical question-answering information, and generates multiple candidate answer information based on the historical question-answering information through a generative large model, predicts the conversion rate of the multiple candidate answer information, obtains the conversion rate information of the candidate answer information, and determines the target answer information for output based on the conversion rate information of the candidate answer information. Therefore, the present disclosure determines the target answer information for output based on the conversion rate information of the candidate answer information, ensures the accuracy of the target answer information, improves the experience in the question-answering process, and improves the conversion rate brought by the target answer information in the intelligent question-answering scenario.

[0037] Figure 2 It is a flowchart of the intelligent question-answering method based on a large model according to the second embodiment of the present disclosure.

[0038] like Figure 2 As shown, in Figure 2 Based on the illustrated embodiment, the intelligent question-answering method based on a large model in the embodiment of the present disclosure may specifically include the following steps:

[0039] S201, obtain historical question and answer information, and generate multiple candidate answer information based on the historical question and answer information through a generative large model.

[0040] For the relevant content of step S201, please refer to the above embodiment and will not be repeated here.

[0041] Optionally, step S102 in the above embodiment of "predicting the conversion rates of multiple candidate reply information and obtaining conversion rate information of the candidate reply information" may specifically include the following steps S202-S203.

[0042] S202: Obtain discrete feature information and text feature information corresponding to the candidate answer information.

[0043] In an embodiment of the present disclosure, by acquiring portrait data of the target object and performing feature extraction on the portrait data, feature information of the target object is obtained, and question and answer feature information is extracted from historical question and answer information, wherein the discrete feature information includes at least feature information of the target object and question and answer feature information.

[0044] For example, in a business Q&A consultation scenario, the target object's characteristic information includes but is not limited to the target object's gender information, consumption level information, and regional information, and the Q&A characteristic information includes but is not limited to Q&A round information, Q&A merchant name information, Q&A timestamp information, Q&A merchant first-level industry information, etc.

[0045] In an embodiment of the present disclosure, by obtaining text feature information of historical question and answer information and candidate reply information, obtaining the search term of the target object, and extracting the text feature information of the search term, the text feature information of the historical question and answer information and candidate reply information and the text feature information of the search term are used as the text feature information corresponding to the candidate reply information.

[0046] S203: predicting the conversion rate of the candidate answer information based on the discrete feature information and the text feature information to obtain conversion rate information of the candidate answer information.

[0047] Alternatively, as Figure 3 As shown, step S203 in the above embodiment "predicting the conversion rate of the candidate answer information based on the discrete feature information and the text feature information to obtain the conversion rate information of the candidate answer information" may specifically include the following steps S301-S304.

[0048] S301: Obtain a first vector representation of discrete feature information.

[0049] S302: Encode the text feature information to obtain a second vector representation of the text feature information.

[0050] S303: Fuse the first vector representation and the second vector representation to obtain fused feature information.

[0051] In the embodiment of the present disclosure, after obtaining the first vector representation and the second vector representation, the first vector representation and the second vector representation can be fused through a fully connected layer to obtain fused feature information.

[0052] S304: predicting the conversion rate of the candidate answer information based on the fused feature information to obtain conversion rate information of the candidate answer information.

[0053] In an embodiment of the present disclosure, after obtaining the fusion feature information, the first conversion probability and the second conversion probability of the candidate reply information can be obtained based on the fusion feature information. The first conversion probability is used to characterize the probability that the target object continues to chat, and the second conversion probability is used to characterize the probability that the target object retains contact information. The first conversion probability and the second conversion probability are weighted to obtain the conversion rate information of the candidate reply information.

[0054] For example, in a business Q&A consultation scenario, the first conversion probability can be understood as the probability that the next statement is issued by the target object after the merchant's customer service completes the response to the current statement. The second conversion probability can be understood as the probability that the next statement is issued by the target object and the target object retains contact information after the merchant's customer service completes the response to the current statement.

[0055] For example, if Figure 4As shown, the following explains the specific process of predicting the conversion rate of candidate answer information based on discrete feature information and text feature information to obtain the conversion rate information of the candidate answer information.

[0056] Optionally, the discrete feature information includes the question and answer merchant name information (username), question and answer timestamp information (hour), and the question and answer merchant first-level industry information (trade name_1st), etc. The feature information of the target object is not repeated here. The discrete feature information is input into the embedding layer embedding respectively, and the embedding output is spliced ​​through the connection layer concat, and the spliced ​​vector is batch normalized (batch norm), and the batch normalized result is input into the fully connected layer (dense) to obtain the first vector representation (sparse embeddings) of the discrete feature information.

[0057] Optionally, the text feature information includes text feature information (context) of historical question and answer information, text feature information (response) of candidate answer information, and text feature information (knowledge) of search terms. The above text feature information is input into the pre-trained base model plato, and the text feature information is encoded through the embedding layer embedding and transformer block in the base model plato to obtain a second vector representation of the text feature information (transformer embeddings).

[0058] Optionally, the first vector representation (sparse embeddings) and the second vector representation (transformer embeddings) are concatenated through the connection layer concat, and the concatenated vector representation is input into the fully connected layer (dense) to achieve the fusion of the first vector representation and the second vector representation to obtain fused feature information, the fused feature information is input into the first conversion probability prediction network to obtain the first conversion probability of the candidate reply information, the fused feature information is input into the first conversion probability prediction network to obtain the second conversion probability of the candidate reply information, the first conversion probability and the second conversion probability are weighted to obtain the conversion rate information of the candidate reply information.

[0059] It should be noted that the present invention uses a generative model that has been pre-trained with a large amount of question-and-answer corpus as the base model plato, inputs text feature information into the base model to obtain the vector output of the base model, that is, outputs the second vector representation, and vector splices the first vector representation and the second vector representation to fuse the two types of features, and connects to a fully connected layer. A multi-objective model architecture of the Entire Space Multi-Task Model (ESMM) type is used to construct a dual-tower model, and the sum of the first conversion probability and the second conversion probability loss is used to perform backpropagation of the model to obtain the final conversion rate information of the candidate answer information.

[0060] Optionally, step S103 in the above embodiment, “determining target reply information for output based on conversion rate information of candidate reply information” may specifically include the following steps S204-S205.

[0061] S204: Acquire quality evaluation information of candidate reply information.

[0062] In an embodiment of the present disclosure, before obtaining quality evaluation information of candidate reply information, after generating multiple candidate reply information based on historical question and answer information through a generative large model, semantic analysis can be performed on the candidate reply information to obtain semantic information of the candidate reply information, and a pre-configured filter word list can be obtained. Based on the filter word list and the semantic information, the multiple candidate reply information can be filtered to obtain the candidate reply information that is finally retained.

[0063] Optionally, before filtering the multiple candidate answer information, the multiple candidate answer information may be pre-processed, for example, by replacing key information such as contact information in the question and answer information with special symbols through keywords and regular expressions.

[0064] Optionally, after filtering multiple candidate reply messages, multiple sentences of a reply message can be merged into one to conform to the corpus format used in pre-training of the generative model.

[0065] It should be noted that the present disclosure does not limit the specific method for obtaining the quality evaluation information of the candidate reply information, and it can be selected according to actual circumstances.

[0066] Optionally, a pre-trained feedback model may be obtained, and quality evaluation information of each candidate response information may be obtained based on the candidate response information through the feedback model.

[0067] For example, the candidate response information can be input into the feedback model to obtain the quality score corresponding to each candidate response information, where the quality score is the quality evaluation information. The higher the quality score, the more consistent the candidate response information is with the business scenario and the values ​​of the target object.

[0068] S205 , acquiring target reply information from a plurality of candidate reply information according to the quality evaluation information and the conversion rate information.

[0069] It should be noted that the present disclosure does not limit the specific method of obtaining target response information from multiple candidate response information based on quality evaluation information and conversion rate information, and the method can be selected according to actual circumstances.

[0070] Optionally, the quality evaluation information and the conversion rate information may be weighted to obtain screening parameters for the candidate response information, and the candidate response information with the largest screening parameter may be selected from multiple candidate response information and output as the target response information.

[0071] Optionally, multiple candidate reply information are sorted from large to small according to quality evaluation information, and the top N candidate reply information are selected, where N is a natural number greater than or equal to 1. From the top N candidate reply information, the candidate reply information with the highest conversion rate information is selected and output as the target reply information.

[0072] It should be noted that the present disclosure does not limit the setting of N, and it can be set according to actual conditions.

[0073] For example, when N is 3, multiple candidate reply information are sorted from large to small according to the quality evaluation information, and the top three candidate reply information are selected from the sorting results of the quality evaluation information of the candidate reply information as candidate reply information 1 (conversion rate information is m1), candidate reply information 5 (conversion rate information is m5), and candidate reply information 7 (conversion rate information is m7). If m5>m1>m7, in this case, the candidate reply information is output as the target reply information.

[0074] Optionally, step S103 in the above embodiment of "determining target reply information for output based on conversion rate information of candidate reply information" may specifically include the following step S206.

[0075] S206, the conversion rate information includes a first conversion probability and a second conversion probability, and the candidate reply information is sorted using at least one of the first conversion probability and the second conversion probability as sorting reference information, and the candidate reply information ranked first is selected and output as the target reply information.

[0076] In an embodiment of the present disclosure, after obtaining the conversion rate information of the candidate reply information, the candidate reply information can be sorted using at least one of the first conversion probability and the second conversion probability as sorting reference information, and the candidate reply information ranked first can be selected and output as the target reply information.

[0077] For example, with the first conversion probability as the sorting reference information, multiple candidate reply information are sorted from large to small according to the first conversion probability, and the candidate reply information ranked first from the sorting results of the first conversion probability of the candidate reply information is selected and output as the target reply information.

[0078] The present disclosure proposes a large-scale model-based intelligent question-answering method, which obtains historical question-answering information and generates multiple candidate answer information based on the historical question-answering information through a generative large-scale model, obtains discrete feature information and text feature information corresponding to the candidate answer information, predicts the conversion rate of the candidate answer information based on the discrete feature information and text feature information, obtains conversion rate information of the candidate answer information, obtains quality evaluation information of the candidate answer information, obtains target answer information from the multiple candidate answer information based on the quality evaluation information and conversion rate information, or the conversion rate information includes a first conversion probability and a second conversion probability, and uses at least one of the first conversion probability and the second conversion probability as sorting reference information to sort the candidate answer information, selects the candidate answer information ranked first, and outputs it as the target answer information. Therefore, the present disclosure predicts the conversion rate of the candidate answer information based on the discrete feature information and text feature information, improves the accuracy and reliability of the conversion rate information, and determines the target answer information for output based on the conversion rate information of the candidate answer information, ensures the accuracy of the target answer information, improves the user experience during the question-answering process, and improves the conversion rate brought by the target answer information in the intelligent question-answering scenario.

[0079] The following uses a business question-and-answer consultation scenario as an application scenario to explain the specific process of the large-model-based intelligent question-and-answer method in the embodiment of the present disclosure.

[0080] For example, if Figure 5 As shown, the intelligent question answering method based on the large model may specifically include the following steps S501-S505.

[0081] S501, obtaining historical question and answer information, and generating multiple candidate answer information based on the historical question and answer information through a business question and answer generation model.

[0082] S502, obtaining semantic information of candidate reply information, filtering multiple candidate reply information based on the merchant's pre-selected filter word list and semantic information, and obtaining the candidate reply information that is finally retained.

[0083] S503: Outputting a floating point score as quality evaluation information of the candidate response information based on the candidate response information through the pre-trained feedback model.

[0084] S504: Acquire conversion rate information of the candidate reply information, wherein the conversion rate information includes a first conversion probability (continuation chat rate) and a second conversion probability (retention rate).

[0085] S505 , performing joint sorting based on the quality evaluation information and the conversion rate information, and obtaining final target reply information from multiple candidate reply information.

[0086] To sum up, the intelligent question-answering method based on a large model proposed in the present disclosure can be applied to business question-answering consulting scenarios. By obtaining the first conversion probability and the second conversion probability, the question-answering rounds of the target object and the merchant in the current scenario are improved. The target object can express his or her questions and needs in a richer language, thereby improving the ability to answer in the current scenario. With the same number of questions and answers, the target object can obtain more clues, thereby improving the conversion rate and operating power in the current scenario, and improving the experience in the question-answering process.

[0087] It should be noted that the information (including but not limited to question and answer information, user device information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals involved in this disclosure are all authorized by the user or fully authorized by all parties, and the source, use and processing of the relevant data comply with relevant laws, regulations and standards, and do not violate public order and good morals.

[0088] According to an embodiment of the present disclosure, the present disclosure also provides an intelligent question-answering device based on a large model, which is used to implement the above-mentioned intelligent question-answering method based on a large model.

[0089] Figure 6 4 is a block diagram of an intelligent question-answering device based on a large model according to an embodiment of the present disclosure.

[0090] like Figure 6 As shown, the intelligent question-answering device 600 based on the large model includes: an acquisition module 601, a prediction module 602 and an output module 603.

[0091] An acquisition module 601 is used to acquire historical question and answer information and generate a plurality of candidate answer information based on the historical question and answer information using a generative large model;

[0092] Prediction module 602, configured to predict the conversion rates of the plurality of candidate answer information and obtain conversion rate information of the candidate answer information;

[0093] The output module 603 is used to determine target reply information for output according to the conversion rate information of the candidate reply information.

[0094] In one embodiment of the present disclosure, the prediction module 602 is also used to: obtain discrete feature information and text feature information corresponding to the candidate reply information; predict the conversion rate of the candidate reply information based on the discrete feature information and the text feature information to obtain the conversion rate information of the candidate reply information.

[0095] In one embodiment of the present disclosure, the prediction module 602 is further used to: obtain a first vector representation of the discrete feature information; encode the text feature information to obtain a second vector representation of the text feature information; fuse the first vector representation and the second vector representation to obtain fused feature information; and predict the conversion rate of the candidate answer information based on the fused feature information to obtain conversion rate information of the candidate answer information.

[0096] In one embodiment of the present disclosure, the prediction module 602 is also used to: obtain a first conversion probability and a second conversion probability of the candidate reply information based on the fused feature information, the first conversion probability being used to characterize the probability that the target object continues to chat, and the second conversion probability being used to characterize the probability that the target object retains contact information; weight the first conversion probability and the second conversion probability to obtain the conversion rate information of the candidate reply information.

[0097] In one embodiment of the present disclosure, the process of acquiring the discrete feature information includes: acquiring portrait data of the target object, and performing feature extraction on the portrait data to obtain feature information of the target object; extracting question and answer feature information from the historical question and answer information, wherein the discrete feature information includes at least the feature information of the target object and the question and answer feature information.

[0098] In one embodiment of the present disclosure, the process of obtaining the text feature information includes: obtaining the text feature information of the historical question and answer information and the candidate answer information; obtaining the search term of the target object, and extracting the text feature information of the search term.

[0099] In one embodiment of the present disclosure, the output module 603 is further used to: obtain quality evaluation information of the candidate response information; and obtain the target response information from the multiple candidate response information based on the quality evaluation information and conversion rate information.

[0100] In one embodiment of the present disclosure, the output module 603 is further used to: weight the quality evaluation information and conversion rate information to obtain the screening parameters of the candidate response information; and select the candidate response information with the largest screening parameter from the multiple candidate response information to output as the target response information.

[0101] In one embodiment of the present disclosure, the output module 603 is also used to: sort the multiple candidate reply information from large to small according to the quality evaluation information, and select the candidate reply information ranked in the top N, where N is a natural number greater than or equal to 1; from the top N candidate reply information, select the candidate reply information with the highest conversion rate information and output it as the target reply information.

[0102] In one embodiment of the present disclosure, the output module 603 is also used for: the conversion rate information includes a first conversion probability and a second conversion probability, and the candidate reply information is sorted using at least one of the first conversion probability and the second conversion probability as sorting reference information, and the candidate reply information ranked first is selected and output as the target reply information.

[0103] In one embodiment of the present disclosure, before obtaining the quality evaluation information of the candidate reply information, the device 600 is also used to: perform semantic analysis on the candidate reply information to obtain the semantic information of the candidate reply information; obtain a pre-configured filter word list; and filter the multiple candidate reply information based on the filter word list and the semantic information to obtain the candidate reply information that is finally retained.

[0104] The intelligent question-and-answer device based on a large model proposed in the present disclosure obtains historical question-and-answer information, and generates multiple candidate answer information based on the historical question-and-answer information through a generative large model, predicts the conversion rate of the multiple candidate answer information, obtains the conversion rate information of the candidate answer information, and determines the target answer information for output based on the conversion rate information of the candidate answer information. Therefore, the present disclosure determines the target answer information for output based on the conversion rate information of the candidate answer information, ensures the accuracy of the target answer information, improves the experience in the question-and-answer process, and improves the conversion rate brought by the target answer information in the intelligent question-and-answer scenario.

[0105] According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0106] Figure 7A schematic block diagram of an example electronic device 700 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0107] like Figure 7 As shown, the device 700 includes a computing unit 701, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. Various programs and data required for the operation of the device 700 can also be stored in the RAM 703. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0108] Various components in device 700 are connected to I / O interface 705, including an input unit 706, such as a keyboard, mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a magnetic disk, optical disk, etc.; and a communication unit 709, such as a network card, modem, wireless communication transceiver, etc. The communication unit 709 allows device 700 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0109] The computing unit 701 can be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 701 performs the various methods and processes described above, such as the intelligent question-answering method based on a large model. For example, in some embodiments, the intelligent question-answering method based on a large model can be implemented as a computer software program that is tangibly contained in a machine-readable medium, such as a storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the intelligent question-answering method based on the large model described above can be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to execute the large model-based intelligent question answering method in any other appropriate manner (for example, by means of firmware).

[0110] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0111] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. Such program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0112] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can 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 can 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.

[0113] To initiate interaction with a user account, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user account; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user account can provide input to the computer. Other types of devices can also be used to initiate interaction with the user account; for example, feedback provided to the user account can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user account can be received in any form (including acoustic input, voice input, or tactile input).

[0114] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user account computer with a graphical user account interface or a web browser through which a user account can interact with embodiments of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by 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), and the Internet.

[0115] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0116] According to an embodiment of the present disclosure, the present disclosure also provides a computer program product, including a computer program, wherein when the computer program is executed by a processor, the steps of the large model-based intelligent question-answering method described in the above embodiment of the present disclosure are implemented.

[0117] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

[0118] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. An intelligent question answering method based on a large model, wherein: The method comprises: Obtain historical question and answer information, and generate multiple candidate answer information based on the historical question and answer information through a generative large model; Predicting conversion rates of the plurality of candidate answer information to obtain conversion rate information of the candidate answer information; Determine target response information for output based on the conversion rate information of the candidate response information; The step of predicting the conversion rates of the plurality of candidate answer information and obtaining the conversion rate information of the candidate answer information includes: Obtaining discrete feature information and text feature information corresponding to the candidate answer information, wherein the discrete feature information at least includes feature information of the target object obtained by extracting features from the target object's portrait data and question-answer feature information extracted from historical question-answer information, and using the text feature information of the historical question-answer information and the candidate answer information and the text feature information of the target object's search term as the text feature information; Predicting the conversion rate of the candidate answer information based on the discrete feature information and the text feature information to obtain conversion rate information of the candidate answer information; The step of predicting the conversion rate of the candidate answer information based on the discrete feature information and the text feature information to obtain the conversion rate information of the candidate answer information includes: Obtaining a first vector representation of the discrete feature information; Encoding the text feature information to obtain a second vector representation of the text feature information; fusing the first vector representation and the second vector representation to obtain fused feature information; Predicting the conversion rate of the candidate answer information based on the fused feature information to obtain conversion rate information of the candidate answer information; The step of predicting the conversion rate of the candidate answer information based on the fused feature information to obtain the conversion rate information of the candidate answer information includes: Obtaining, based on the fused feature information, a first conversion probability and a second conversion probability of the candidate reply information, wherein the first conversion probability is used to represent the probability that the target object continues chatting, and the second conversion probability is used to represent the probability that the target object retains contact information; The first conversion probability and the second conversion probability are weighted to obtain conversion rate information of the candidate answer information.

2. The method according to claim 1, wherein The process of obtaining the discrete feature information includes: Acquire image data of the target object, and perform feature extraction on the image data to obtain feature information of the target object; Question and answer feature information is extracted from the historical question and answer information, wherein the discrete feature information at least includes feature information of the target object and question and answer feature information.

3. The method according to claim 1, wherein The process of obtaining the text feature information includes: Obtaining text feature information of the historical question and answer information and the candidate answer information; Obtain a search term for a target object and extract text feature information of the search term.

4. The method according to any one of claims 1 to 3, wherein The determining target reply information for output based on the conversion rate information of the candidate reply information includes: Obtaining quality evaluation information of the candidate response information; The target answer information is obtained from the plurality of candidate answer information according to the quality evaluation information and the conversion rate information.

5. The method according to claim 4, wherein The acquiring the target response information from the plurality of candidate response information according to the quality evaluation information and the conversion rate information includes: Weighting the quality evaluation information and the conversion rate information to obtain screening parameters for the candidate reply information; From the multiple candidate response information, the candidate response information with the largest screening parameter is selected and output as the target response information.

6. The method according to claim 4, wherein: The acquiring the target response information from the plurality of candidate response information according to the quality evaluation information and the conversion rate information includes: Sort the multiple candidate reply information from largest to smallest according to the quality evaluation information, and select the top N candidate reply information, where N is a natural number greater than or equal to 1; From the first N candidate answer information, the candidate answer information with the highest conversion rate information is selected and output as the target answer information.

7. The method according to claim 1, wherein The determining target reply information for output based on the conversion rate information of the candidate reply information includes: The conversion rate information includes a first conversion probability and a second conversion probability. At least one of the first conversion probability and the second conversion probability is used as sorting reference information to sort the candidate reply information, and the candidate reply information ranked first is selected and output as the target reply information.

8. The method according to claim 4, wherein: Before obtaining the quality evaluation information of the candidate reply information, the method further includes: Performing semantic analysis on the candidate reply information to obtain semantic information of the candidate reply information; Get a list of pre-configured filter words; Based on the filter word list and the semantic information, the plurality of candidate answer information are filtered to obtain the candidate answer information that is finally retained.

9. An intelligent question-answering device based on a large model, wherein: The device comprises: An acquisition module is used to acquire historical question and answer information and generate multiple candidate answer information based on the historical question and answer information through a generative large model; A prediction module, configured to predict conversion rates of the plurality of candidate answer information and obtain conversion rate information of the candidate answer information; An output module, configured to determine target response information for output based on the conversion rate information of the candidate response information; Among them, the prediction module is used to: Obtaining discrete feature information and text feature information corresponding to the candidate answer information, wherein the discrete feature information at least includes feature information of the target object obtained by extracting features from the target object's portrait data and question-answer feature information extracted from historical question-answer information, and using the text feature information of the historical question-answer information and the candidate answer information and the text feature information of the target object's search term as the text feature information; Predicting the conversion rate of the candidate answer information based on the discrete feature information and the text feature information to obtain conversion rate information of the candidate answer information; Among them, the prediction module is used to Obtaining a first vector representation of the discrete feature information; Encoding the text feature information to obtain a second vector representation of the text feature information; fusing the first vector representation and the second vector representation to obtain fused feature information; Predicting the conversion rate of the candidate answer information based on the fused feature information to obtain conversion rate information of the candidate answer information; Among them, the prediction module is used to: Obtaining, based on the fused feature information, a first conversion probability and a second conversion probability of the candidate reply information, wherein the first conversion probability is used to represent the probability that the target object continues chatting, and the second conversion probability is used to represent the probability that the target object retains contact information; The first conversion probability and the second conversion probability are weighted to obtain conversion rate information of the candidate answer information.

10. An electronic device, characterized in that: including processor and memory; The processor reads the executable program code stored in the memory to run a program corresponding to the executable program code, so as to implement the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

12. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 8.

Images