Conversation data set construction system for training user item preference recommendation

Abstract

The invention provides a dialogue data set construction system for training user project preference recommendation. The dialogue data set construction system comprises a text dialogue generation module, a voice dialogue generation module and a dialogue data quality evaluation module. Through a text dialogue generation module, a voice dialogue generation module and a dialogue data quality evaluation module, text-based dialogues and voice-based dialogues are constructed for a user side and a virtual assistant side, and dialogue data quality evaluation is generated, so that text dialogue content is generated, and then corresponding roles are selected according to user conditions. The best role is matched, and finally, the generation quality is subjected to all-around evaluation. Therefore, a brand new dialogue thought is provided, and the recommendation precision of the dialogue process is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence dialogue recommendation technology, and in particular to a dialogue data set construction system for training user item preference recommendation, an electronic device and a readable storage medium. Background Art

[0002] Conversational Recommendation Systems (CRS) have recently gained great attention in the field of Recommender Systems (RS). The natural advantages of CRS, such as the ability to obtain user preferences through interactive dialogues and provide intuitive explanations for recommendations, are in line with the growing trend of explainable AI. Most of the research on CRS to date has focused on text-based conversations between users and virtual assistants (Text-based Convertional Recommendation Systems, TCRS). However, TCRS is not user-friendly for some users, such as those with visual impairments or limited writing and reading abilities.

[0003] Compared with text, voice conversations are more direct, natural, convenient, and time-saving. Using voice input in CRS (i.e., Voice-based Convertional Recommendation Systems, VCRS) can bring a series of benefits. First, the user experience is more direct, attractive, and efficient because users are less likely to be distracted or lost browsing products. With VCRS, users can ask exactly what they want and place orders conveniently. Second, VCRS are more conversational in nature and allow for the inclusion of objective deep features and subjective features in the user's query. Furthermore, voice input can convey more information than text input, such as user age, gender, and accent, which enables VCRS to provide more accurate and transparent recommendations. Finally, VCRS are more user-friendly and easy to use, especially for individuals who have difficulty writing and reading.

[0004] In addition to providing a user-friendly and convenient way, voice conversations also convey more information (such as age, gender, accent, and emotional state) than plain text conversations. In addition, extracting this auxiliary information from speech is necessary and more reliable for two reasons:

[0005] 1) Some auxiliary information can only come from voice, such as accent and emotional state. It is difficult to obtain such detailed and dynamic information from user registration behind the backend system.

[0006] 2) Although text can obtain some auxiliary information (e.g., gender, age), there are usually problems with missing data and false information, i.e., user registration information is randomly entered or has restricted access. In contrast, human voice is honest in conveying such information, especially when conducting voice-based conversations.

[0007] Therefore, in order to effectively utilize the features of text and voice data, it is necessary to construct a dialogue dataset suitable for training and testing user item preference recommendation models to ensure the recommendation accuracy of the model.

[0008] It can be seen that the shortcomings of the prior art and the reasons for these shortcomings are:

[0009] First, text-based conversational recommendation systems face the problems of difficulty in understanding user intent and complexity in conversation management, especially in dealing with ambiguous expressions and maintaining consistency in multi-round conversations.

[0010] Second, the accuracy of personalized recommendations and user experience are affected by insufficient data, slow system response, and sentiment recognition errors, which may lead to poor recommendation results and reduced user satisfaction. Summary of the invention

[0011] In view of the above problems, the present invention is proposed to provide a conversation dataset construction system, an electronic device and a readable storage medium for training user item preference recommendation, which overcome the above problems or at least partially solve the above problems.

[0012] The present invention provides a conversation data set construction system for training user item preference recommendation, comprising:

[0013] A text dialogue generation module is used to generate a dialogue template for the user side and a dialogue template for the virtual assistant side according to the project theme; the dialogue template includes a question template and a reply template; and select a target dialogue template for each round of dialogue between the user side and the virtual assistant side from the dialogue template; and analyze the user-project interaction data to determine the importance ranking result of each dialogue feature, and according to the dialogue feature importance ranking result from high to low, use the target dialogue template and dialogue features to generate at least one round of questions and obtain the real item feature value of the dialogue feature, so as to obtain the text dialogue between the user side and the virtual assistant side;

[0014] A speech dialogue generation module, which is used to convert the text dialogue between the user and the virtual assistant into a speech dialogue using an end-to-end text-to-speech model;

[0015] The conversation data quality assessment module is used to assess the quality of text conversations and voice conversations between the user and virtual assistant.

[0016] Optionally, each dialog template is configured with a preset tag;

[0017] The text dialogue generation module is also used for:

[0018] In each round of conversation, a conversation template is randomly selected from the user end or the virtual assistant end, and a corresponding conversation template is selected from the other end according to a predefined tag compatibility mode to form a conversation template matched in each round of conversation; the predefined tag compatibility mode is compatible with conversation templates with the same preset tags in the user end and the virtual assistant end.

[0019] Optionally, the inquiry template corresponds to different questioning methods, including a selection-based inquiry template and a query-based inquiry template;

[0020] The text dialogue generation module is also used to select inquiry templates with different questioning methods for the virtual assistant end according to the current dialogue stage; if the current dialogue stage is the question-asking stage, a selection-based inquiry template is selected for the virtual assistant end.

[0021] Optionally, the text dialogue generation module is further used for:

[0022] Clean and feature-engineer the user-project interaction data to obtain conversation features;

[0023] The LightGBM model based on decision tree is used to analyze the importance of each dialogue feature and obtain the importance score of each dialogue feature;

[0024] The conversation features are ranked according to their importance scores to obtain the importance ranking results of the conversation features.

[0025] Optionally, the text dialogue generation module is further used for:

[0026] Determine the number of rounds of questions according to the questioning method of the inquiry template selected for the virtual assistant;

[0027] If the inquiry template selected for the virtual assistant end is a query-based inquiry template, the number of rounds of questioning is one round;

[0028] If the inquiry template selected for the virtual assistant is a selection-based inquiry template, the number of question rounds is determined according to the time point when the feature value of the real item is successfully obtained.

[0029] Optionally, the voice dialogue generation module is used to:

[0030] A first end-to-end text-to-speech model is used to convert a text conversation on the virtual assistant end into a speech conversation; the first end-to-end text-to-speech model is trained using an LJ Speech dataset; the LJ Speech dataset contains different short audio clips of a single speaker.

[0031] Optionally, the voice dialogue generation module is used to:

[0032] A second end-to-end text-to-speech model is used to convert text conversations on the user side into voice conversations that match basic personal information of different users; the second end-to-end text-to-speech model is trained using a Vctk dataset; the Vctk dataset contains voice data of different speakers with different accents.

[0033] Optionally, the conversation data quality assessment module is used to:

[0034] The fine-grained dialogue evaluation index is used to measure the dialogue quality of the text dialogue between the user end and the virtual assistant end, and the same evaluation on multiple real-world human dialogue data sets is used as a benchmark; the dimensions of the fine-grained dialogue evaluation index include comprehension, relevance, coherence, interactivity, interest, diversity, naturalness, accuracy, informativeness, conciseness, fluency, consistency, trust, politeness, supportiveness, innovation, emotional expression and inquiry;

[0035] For the text conversation fragments extracted from the text conversation between the user end and the virtual assistant end, an end-to-end text-to-speech model is used to generate the voice conversation between the user end and the virtual assistant end, and a two-stage TTS model is used to generate voice data. Based on the user survey results, the quality of the generated audio fragments is evaluated and compared in terms of audio naturalness, speaker age recognition accuracy, and speaker gender recognition accuracy.

[0036] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the electronic device executes the computer program, it loads a conversation data set construction system for training user item preference recommendations as described in any one of the embodiments of the present invention.

[0037] The present invention also provides a readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the system for constructing a conversation data set for training user item preference recommendations as described in any one of the embodiments of the present invention is loaded.

[0038] The present invention includes the following advantages:

[0039] The conversation data set construction system for training user item preference recommendation of the present invention realizes the generation of text conversation content by constructing text-based conversations and voice-based conversations for the user end and the virtual assistant end, and generating conversation data quality evaluation, and then selects the corresponding role according to the user situation to match the best role, and finally aligns the generation quality for a comprehensive evaluation. This provides a new conversation idea and improves the recommendation accuracy of the conversation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A dialog dataset construction system for training user item preference recommendations is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] Reference Figure 1 , shows a structural block diagram of a conversation dataset construction system for training user item preference recommendation provided in an embodiment of the present invention, which may specifically include the following modules:

[0043] A text dialogue generation module is used to generate a dialogue template for the user side and a dialogue template for the virtual assistant side according to the project theme; the dialogue template includes a question template and a reply template; and select a target dialogue template for each round of dialogue between the user side and the virtual assistant side from the dialogue template; and analyze the user-project interaction data to determine the importance ranking result of each dialogue feature, and according to the dialogue feature importance ranking result from high to low, use the target dialogue template and dialogue features to generate at least one round of questions and obtain the real item feature value of the dialogue feature, so as to obtain the text dialogue between the user side and the virtual assistant side;

[0044] A speech dialogue generation module, which is used to convert the text dialogue between the user and the virtual assistant into a speech dialogue using an end-to-end text-to-speech model;

[0045] The conversation data quality assessment module is used to assess the quality of text conversations and voice conversations between the user and virtual assistant.

[0046] In one embodiment of the present invention, each dialog template is configured with a preset tag;

[0047] The text dialogue generation module is also used for:

[0048] In each round of conversation, a conversation template is randomly selected from the user end or the virtual assistant end, and a corresponding conversation template is selected from the other end according to a predefined tag compatibility mode to form a conversation template matched in each round of conversation; the predefined tag compatibility mode is compatible with conversation templates with the same preset tags in the user end and the virtual assistant end.

[0049] In one embodiment of the present invention, the inquiry template corresponds to different questioning methods, including a selection-based inquiry template and a query-based inquiry template;

[0050] The text dialogue generation module is also used to select inquiry templates with different questioning methods for the virtual assistant end according to the current dialogue stage; if the current dialogue stage is the question-asking stage, a selection-based inquiry template is selected for the virtual assistant end.

[0051] In one embodiment of the present invention, the text dialogue generation module is further used for:

[0052] Clean and feature-engineer the user-project interaction data to obtain conversation features;

[0053] The LightGBM model based on decision tree is used to analyze the importance of each dialogue feature and obtain the importance score of each dialogue feature;

[0054] The conversation features are ranked according to their importance scores to obtain the importance ranking results of the conversation features.

[0055] In one embodiment of the present invention, the text dialogue generation module is further used for:

[0056] Determine the number of rounds of questions according to the questioning method of the inquiry template selected for the virtual assistant;

[0057] If the inquiry template selected for the virtual assistant end is a query-based inquiry template, the number of rounds of questioning is one round;

[0058] If the inquiry template selected for the virtual assistant is a selection-based inquiry template, the number of question rounds is determined according to the time point when the feature value of the real item is successfully obtained.

[0059] In one embodiment of the present invention, the voice dialogue generation module is used to:

[0060] A first end-to-end text-to-speech model is used to convert a text conversation on the virtual assistant end into a speech conversation; the first end-to-end text-to-speech model is trained using an LJ Speech dataset; the LJ Speech dataset contains different short audio clips of a single speaker.

[0061] In one embodiment of the present invention, the voice dialogue generation module is used to:

[0062] A second end-to-end text-to-speech model is used to convert text conversations on the user side into voice conversations that match basic personal information of different users; the second end-to-end text-to-speech model is trained using a Vctk dataset; the Vctk dataset contains voice data of different speakers with different accents.

[0063] In one embodiment of the present invention, the conversation data quality assessment module is used to:

[0064] The fine-grained dialogue evaluation index is used to measure the dialogue quality of the text dialogue between the user end and the virtual assistant end, and the same evaluation on multiple real-world human dialogue data sets is used as a benchmark; the dimensions of the fine-grained dialogue evaluation index include comprehension, relevance, coherence, interactivity, interest, diversity, naturalness, accuracy, informativeness, conciseness, fluency, consistency, trust, politeness, supportiveness, innovation, emotional expression and inquiry;

[0065] For the text conversation fragments extracted from the text conversation between the user end and the virtual assistant end, an end-to-end text-to-speech model is used to generate the voice conversation between the user end and the virtual assistant end, and a two-stage TTS model is used to generate voice data. Based on the user survey results, the quality of the generated audio fragments is evaluated and compared in terms of audio naturalness, speaker age recognition accuracy, and speaker gender recognition accuracy.

[0066] In the present invention, for each user-item interaction record, the goal is to simulate multiple rounds of conversations by asking the user about their preferences for features related to the item until the item is successfully recommended. The application scenarios of the conversation recommendation of the present invention are very broad. The goal of system design is to capture the user's preference information during the interaction with any type of item or topic (such as services, content, experience, etc.). Therefore, the conversation recommendation scheme is suitable for a variety of scenarios where it is necessary to understand user preferences and make personalized recommendations. In the process of introducing the scheme, the present invention uses item features as an example to illustrate how the system captures user preferences through multiple rounds of conversations, but the conversation recommendation of the present invention is not limited to the recommendation commodity scenario.

[0067] In the process of generating the text dialogue between the user and the virtual assistant, the dialogue templates for the user and the virtual assistant are generated respectively. In order to ensure the naturalness and diversity of the dialogue, different inquiry and response templates need to be generated.

[0068] Among them, the user-side template is mainly used to simulate the user's behavior and reaction in the conversation. This template includes the types of questions that the user may ask, the way of answering, and the choice preferences in different situations. Its purpose is to provide the system with a variety of user interaction modes so that when testing and training the model, it can simulate the conversation situation as close to the real user as possible. The virtual assistant-side template is used to define the response mode of the system when interacting with the user. This template includes the way the virtual assistant answers the user's questions, the suggestions provided, and how to adjust the next inquiry or suggestion based on the user's feedback. Its purpose is to ensure the logic, accuracy and personalization of the virtual assistant's response. The difference between the two lies in the role positioning and interaction goals: the user-side template focuses on simulating the diverse behaviors of users, while the virtual assistant-side template focuses on optimizing the system's response. By combining these two different templates, the rationality and richness of the dialogue generation process can be ensured.

[0069] Especially for virtual assistants, when given each item feature, the present invention will change the way of asking questions, either directly asking the user's favorite item features (based on the inquiry, tag identifier tag-id: 00), or providing single or multiple choices (based on the selection, tag identifier tag-id: 01).

[0070] In addition, the present invention also generates templates for virtual assistants and users about item recommendations and opening / closing conversations. All these templates under different tag-ids can be generated by the large language model (LLM) ChatGPT released by OpenAI. The training of ChatGPT largely follows InstrucGPT, which uses supervised training of human expert labels for question-answering dialogues and reinforcement learning through expert-driven reward models.

[0071] After generating the dialogue templates for the user side and the virtual assistant side, it is necessary to select the appropriate template for each dialogue between the virtual assistant and the user. Due to the diversity of templates, it is inevitable that there will be dialogues with mismatched questions and answers. In order to avoid such embarrassing situations and increase the coherence of the dialogue, the present invention designs the following strategies:

[0072] Specifically, the present invention first adds a tag tag-id to each template of the virtual assistant and the user, and defines compatible matching patterns for these tags. Only templates with the same tag-id on the virtual assistant side and the user side are compatible. Therefore, in each round of the conversation, the present invention will randomly sample a template from the virtual assistant side and the user side respectively under the guidance of the predefined compatible pattern. Therefore, this retains the rationality of the generated conversation.

[0073] Specifically, in the question-posing stage, the present invention provides a greater probability for the virtual assistant to sample a selection-based template (tag-id: 01) rather than a query-based template (tag-id: 00). This is because an intuitive fact is that users usually do not have a clear intention at the beginning when interacting with CRS. In other words, CRS needs to trigger user preferences through multiple rounds of dialogue.

[0074] After providing the selected template, it is necessary to determine the best question, that is, it is necessary to select features from the conversation so that the system can quickly identify user preferences and ultimately make better recommendations. For example, in the commodity recommendation scenario, it is to determine the features of the item being asked, and in the functional service scenario, it is to determine the functional features. The present invention designs the following strategies:

[0075] Specifically, it is necessary to determine the order of the feature types being asked, because different types of features may have different impacts on the user's final decision. If the feature types with greater impact can be asked earlier, the user's preferences can be inferred faster. Therefore, the present invention uses the decision tree-based method LightGBM to calculate the importance of different feature types. Feature types with higher importance values ​​obtained by LightGBM are closer to the root, indicating that they have a greater impact on the user's final decision, so they should be asked at an earlier stage. The resulting decision tree will eventually participate in the text-based dialogue generation process to determine the order of the feature types asked.

[0076] When determining the order of feature types, this paper uses LightGBM, a gradient boosting framework based on decision trees, to calculate the importance of each feature. The specific steps are as follows:

[0077] 1. Data preprocessing: First, the user interaction data will be cleaned and feature engineered to ensure the quality and consistency of the features.

[0078] 2. Training model: The processed data is then input into the LightGBM model for training. During the training process, the model will build a decision tree based on the importance of the features and output the importance score of each feature.

[0079] 3. Feature sorting: According to the importance score output by the LightGBM model, the features are sorted from high to low. Features with higher scores usually have a greater impact on user decisions, and these features will be closer to the root in the decision tree, indicating that they contribute more to the model prediction.

[0080] 4. Dialogue optimization: Using the ranking results of these feature importance scores, users are asked questions about high-importance features first in the dialogue system, so as to capture user preferences more quickly and accurately and optimize the recommendation effect.

[0081] Specifically, the present invention needs to determine the number of turns for each feature type inquired. If the selected virtual assistant template is query-based, the present invention can directly obtain the real item feature value of the inquired feature type from the user in just one turn. If it is selection-based, the number of turns depends on when the real feature value can be sampled. In order to ensure the rationality of the dialogue, the present invention follows the interactive distribution, that is, the feature value of the feature type inquired when the feature value is sampled based on the selected template.

[0082] After generating the text dialogue between the user and the virtual assistant, the generated text dialogue can be converted into voice dialogue to ensure that the system can generate realistic voice interaction scenarios, thereby better simulating the user experience. Through advanced technical means, the user experience and the ability of the dialogue system can be improved, while promoting the widespread application and development of technology.

[0083] Specifically, the generation of text-to-speech dialogue is realized. The present invention uses an end-to-end text-to-speech model VITS to complete the generation of text-to-speech dialogue. For the virtual assistant end, a VITS model trained by the LJ Speech dataset is used, and its LJ Speech dataset can contain short audio clips of a single speaker. For the user end, a VITS model trained by the Vctk dataset is used, and its Vctk dataset can contain speech data of 109 English speakers with different accents. For each text-based dialogue, a suitable speaker is matched according to the user's gender and age so that it can be converted into a voice dialogue.

[0084] The quality of the constructed conversation datasets can be evaluated separately for the generated text conversations and voice conversations.

[0085] In order to automatically and systematically measure the quality of generated text-based dialogues, the present invention adopts the fine-grained evaluation of dialogue (FED) metric, which is an automatic evaluation metric using the large-scale pre-trained model DialoGPT. Using the FED metric has three advantages: 1) it does not rely on real responses, 2) it does not require any training data, and 3) it can measure fine-grained dialogue quality (such as correctness, comprehensibility, and fun) and the overall dialogue level (such as coherence, consistency, and diversity) at both the turn level. In order to ensure the effectiveness of the dataset constructed by the present invention, the same evaluation of three real-world human dialogue datasets (i.e., ReDial, OpenDialKG, and INSPIRED) is used as a benchmark.

[0086] Specifically, each time we use a human conversation dataset as a benchmark (such as ReDial), and calculate the performance gap between the benchmark and the other two datasets (such as OpenDialKG and INSPIRED). Generally speaking, the smaller the performance gap with the benchmark, the better the quality of the corresponding dataset.

[0087] Specifically, the results obtained by FED in 18 aspects at the turn and dialogue levels (comprehension, relevance, coherence, interactivity, interest, diversity, naturalness, accuracy, informativeness, conciseness, fluency, consistency, trust, politeness, supportiveness, innovation, emotional expression, and inquiry) are averaged as the final performance score: "Average" refers to the average score of the three benchmark data sets; "Gap-R" represents the absolute performance gap between the other two models.

[0088] In order to evaluate the quality of the generated voice dialogues, the present invention compares the voice data generated by end-to-end VITS with the voice data generated by three two-stage TTS models (i.e., Text2Mel+Vocoders): Tacotron 2+WaveGAN (TacoW), Transformer+WaveGAN (TranW), and FastSpeech 2+WaveGAN (FastW).

[0089] Specifically, we first randomly extract 10 text clips from the text-based conversations generated on the selected dataset. Then, we use VITS and three baseline TTS models to generate the corresponding audio clips. Finally, we get 20 groups of audio clips, each of which is associated with a same text clip and four audio clips generated by four TTS models.

[0090] Specifically, based on user research, the quality of the four synthesized audio clips in each group is evaluated from three different perspectives:

[0091] (1) Obtaining mean opinion score (MOS), i.e., the naturalness of the audio clip, is classified into 1-5 levels (bad, poor, fair, good, excellent);

[0092] (2) Use different age options to determine the speaker's age;

[0093] (3) Use three options to determine whether the speaker is male, female, or other.

[0094] The present invention recruited multiple participants to answer the 20 sets of questions one by one. In order to conduct a fair evaluation without introducing personal bias, four audio clips generated by four TTS models were randomly selected in each group to avoid some participants from scoring some audio clips too high or too low.

[0095] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored on the memory, wherein the processor executes the computer program to load a conversation dataset construction system for training user item preference recommendations as described in any one of the embodiments of the present invention.

[0096] Specifically, the electronic device includes: a memory and a processor, the memory and the processor are connected via a bus communication, the memory stores a computer program, and the computer program can be run on the processor, thereby loading the conversation data set construction system for training user item preference recommendation as described in any one of the first aspects of the embodiments of the present invention.

[0097] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0098] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0099] Based on the same inventive concept, another embodiment of the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon. When the computer program / instruction is executed by a processor, the system for constructing a conversation dataset for training user item preference recommendations as described in any one of the first aspects of the embodiments of the present invention is loaded.

[0100] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0101] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, devices, electronic devices, storage media or computer program products. Therefore, embodiments of the present invention may take the form of complete hardware embodiments, complete software embodiments, or embodiments combining software and hardware. Moreover, embodiments of the present invention may take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CDROM, optical storage, etc.) containing computer-usable program codes.

[0102] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0103] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0104] The above is a detailed introduction to a conversation data set construction system, electronic device and readable storage medium for training user item preference recommendations provided by the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and its core idea of ​​the present application; at the same time, for those of ordinary skill in the art, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present application. The above embodiments are only preferred embodiments cited to fully illustrate the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or changes made by those skilled in the art on the basis of the present invention are all within the protection scope of the present invention.

Claims

1. A conversation dataset construction system for training user item preference recommendation, characterized in that: The system comprises: A text dialogue generation module is used to generate a dialogue template for the user side and a dialogue template for the virtual assistant side according to the project theme; the dialogue template includes a question template and a reply template; and select a target dialogue template for each round of dialogue between the user side and the virtual assistant side from the dialogue template; and analyze the user-project interaction data to determine the importance ranking result of each dialogue feature, and according to the dialogue feature importance ranking result from high to low, use the target dialogue template and dialogue features to generate at least one round of questions and obtain the real item feature value of the dialogue feature, so as to obtain the text dialogue between the user side and the virtual assistant side; A speech dialogue generation module, which is used to convert the text dialogue between the user and the virtual assistant into a speech dialogue using an end-to-end text-to-speech model; The conversation data quality assessment module is used to assess the quality of text conversations and voice conversations between the user and virtual assistant.

2. The system for constructing a conversation dataset for training user item preference recommendation according to claim 1, characterized in that: Each conversation template is configured with a preset label; The text dialogue generation module is also used for: In each round of dialogue, a dialogue template is randomly selected from the user end or the virtual assistant end, and a corresponding dialogue template is selected from the other end according to a predefined tag compatibility mode to form a dialogue template matched in each round of dialogue; The predefined tag compatibility mode is compatible with the dialog templates with the same preset tags in the user terminal and the virtual assistant terminal.

3. The system for constructing a conversation dataset for training user item preference recommendation according to claim 1, characterized in that: The inquiry templates correspond to different questioning methods, including selection-based inquiry templates and query-based inquiry templates; The text dialogue generation module is also used to select inquiry templates with different questioning methods for the virtual assistant end according to the current dialogue stage; if the current dialogue stage is the question-asking stage, a selection-based inquiry template is selected for the virtual assistant end.

4. The system for constructing a conversation dataset for training user item preference recommendation according to claim 1, characterized in that: The text dialogue generation module is also used for: Clean and feature-engineer the user-project interaction data to obtain conversation features; The LightGBM model based on decision tree is used to analyze the importance of each dialogue feature and obtain the importance score of each dialogue feature; The conversation features are ranked according to their importance scores to obtain the importance ranking results of the conversation features.

5. The system for constructing a conversation dataset for training user item preference recommendation according to claim 3, characterized in that: The text dialogue generation module is also used for: Determine the number of rounds of questions according to the questioning method of the inquiry template selected for the virtual assistant; If the inquiry template selected for the virtual assistant end is a query-based inquiry template, the number of rounds of questioning is one round; If the inquiry template selected for the virtual assistant is a selection-based inquiry template, the number of question rounds is determined according to the time point when the feature value of the real item is successfully obtained.

6. The system for constructing a conversation dataset for training user item preference recommendation according to claim 1, characterized in that: The voice dialogue generation module is used for: A first end-to-end text-to-speech model is used to convert a text conversation on the virtual assistant end into a speech conversation; the first end-to-end text-to-speech model is trained using an LJ Speech dataset; the LJ Speech dataset contains different short audio clips of a single speaker.

7. The system for constructing a conversation dataset for training user item preference recommendation according to claim 1, characterized in that: The voice dialogue generation module is used for: A second end-to-end text-to-speech model is used to convert text conversations on the user side into voice conversations that match basic personal information of different users; the second end-to-end text-to-speech model is trained using a Vctk dataset; the Vctk dataset contains voice data of different speakers with different accents.

8. The system for constructing a conversation dataset for training user item preference recommendation according to claim 1, characterized in that: The dialogue data quality assessment module is used to: The fine-grained dialogue evaluation index is used to measure the dialogue quality of the text dialogue between the user end and the virtual assistant end, and the same evaluation on multiple real-world human dialogue data sets is used as a benchmark; the dimensions of the fine-grained dialogue evaluation index include comprehension, relevance, coherence, interactivity, interest, diversity, naturalness, accuracy, informativeness, conciseness, fluency, consistency, trust, politeness, supportiveness, innovation, emotional expression and inquiry; For the text conversation fragments extracted from the text conversation between the user end and the virtual assistant end, an end-to-end text-to-speech model is used to generate the voice conversation between the user end and the virtual assistant end, and a two-stage TTS model is used to generate voice data. Based on the user survey results, the quality of the generated audio fragments is evaluated and compared in terms of audio naturalness, speaker age recognition accuracy, and speaker gender recognition accuracy.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the electronic device executes the computer program, it loads the conversation data set construction system for training user item preference recommendation as described in any one of claims 1 to 8.

10. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the system for constructing a conversation data set for training user item preference recommendation as described in any one of claims 1 to 8 is loaded.

Images