Natural Language Processing
M. J. Nasri-Lowshani; J. Salimi Sartakhti; H. Ebrahimpour-Komole
Abstract
Background and Objectives: Developing efficient task-oriented dialogue systems capable of handling multilingual interactions is a growing area of research in natural language processing (NLP). In this paper, we propose SenSimpleDS, a deep reinforcement learning-based joint task-oriented dialogue system, ...
Read More
Background and Objectives: Developing efficient task-oriented dialogue systems capable of handling multilingual interactions is a growing area of research in natural language processing (NLP). In this paper, we propose SenSimpleDS, a deep reinforcement learning-based joint task-oriented dialogue system, designed for multilingual conversations.Methods: The system utilizes a deep Q-network and the SBERT model to represent the dialogue environment. We introduce two variants, SenSimpleDS+ and SenSimpleDS-NSP, which incorporate modifications in the ε-greedy method and leverage next sequence prediction (NSP) using BERT to refine the reward function. These methods are evaluated on datasets in English, Persian, Spanish, and German, and compared with baseline methods such as SimpleDS and SCGSimpleDS.Results: Our experimental results demonstrate that the proposed methods outperform the baselines in terms of average collected rewards, requiring fewer learning steps to achieve optimal dialogue policies. Notably, the incorporation of NSP significantly improves performance by optimizing reward collection. The multilingual SenSimpleDS further showcases the system’s ability to function across languages using a random forest classifier for language detection and MPNet for environment construction. In addition to system evaluations, we introduce a new Persian dataset for task-oriented dialogue in the restaurant domain, expanding the resources available for developing dialogue systems in low-resource languages.Conclusion: SenSimpleDS, a deep reinforcement learning-based joint task-oriented dialogue system, demonstrates superior performance over baseline methods by leveraging deep Q-networks, SBERT. The integration of next sequence prediction (NSP) significantly enhances reward optimization, enabling faster convergence to optimal dialogue policies. This work establishes a foundation for future research in multilingual dialogue systems, with potential applications across diverse service domains.
Natural Language Processing
M. Khazeni; M. Heydari; A. Albadvi
Abstract
Background and Objectives: The lack of a suitable tool for the analysis of conversational texts in Persian language has made various analyzes of these texts, including Sentiment Analysis, difficult. In this research, it has we tried to make the understanding of these texts easier for the machine by providing ...
Read More
Background and Objectives: The lack of a suitable tool for the analysis of conversational texts in Persian language has made various analyzes of these texts, including Sentiment Analysis, difficult. In this research, it has we tried to make the understanding of these texts easier for the machine by providing PSC, Persian Slang Convertor, a tool for converting conversational texts into formal ones, and by using the most up-to-date and best deep learning methods along with the PSC, the sentiment learning of short Persian language texts for the machine in a better way.Methods: Be made More than 10 million unlabeled texts from various social networks and movie subtitles (as dialogue texts) and about 10 million news texts (as official texts) have been used for training unsupervised models and formal implementation of the tool. 60,000 texts from the comments of Instagram social network users with positive, negative, and neutral labels are considered as supervised data for training the emotion classification model of short texts. The latest methods such as LSTM, CNN, BERT, ELMo, and deep processing techniques such as learning rate decay, regularization, and dropout have been used. LSTM has been utilized in the research, and the best accuracy has been achieved using this method.Results: Using the official tool, 57% of the words of the corpus of conversation were converted. Finally, by using the formalizer, FastText model and deep LSTM network, the accuracy of 81.91 was obtained on the test data.Conclusion: In this research, an attempt was made to pre-train models using unlabeled data, and in some cases, existing pre-trained models such as ParsBERT were used. Then, a model was implemented to classify the Sentiment of Persian short texts using labeled data.
Natural Language Processing
Y. Saffari; J. Salimi Sartakhti
Abstract
Background and Objectives: Most of the recent dialogue policy learning methods are based on reinforcement learning (RL). However, the basic RL algorithms like deep Q-network, have drawbacks in environments with large state and action spaces such as dialogue systems. Most of the policy-based ...
Read More
Background and Objectives: Most of the recent dialogue policy learning methods are based on reinforcement learning (RL). However, the basic RL algorithms like deep Q-network, have drawbacks in environments with large state and action spaces such as dialogue systems. Most of the policy-based methods are slow, cause of the estimating of the action value using the computation of the sum of the discounted rewards for each action. In value-based RL methods, function approximation errors lead to overestimation in value estimation and finally suboptimal policies. There are works that try to resolve the mentioned problems using combining RL methods, but most of them were applied in the game environments, or they just focused on combining DQN variants. This paper for the first time presents a new method that combines actor-critic and double DQN named Double Actor-Critic (DAC), in the dialogue system, which significantly improves the stability, speed, and performance of dialogue policy learning. Methods: In the actor critic to overcome the slow learning of normal DQN, the critic unit approximates the value function and evaluates the quality of the policy used by the actor, which means that the actor can learn the policy faster. Moreover, to overcome the overestimation issue of DQN, double DQN is employed. Finally, to have a smoother update, a heuristic loss is introduced that chooses the minimum loss of actor-critic and double DQN. Results: Experiments in a movie ticket booking task show that the proposed method has more stable learning without drop after overestimation and can reach the threshold of learning in fewer episodes of learning. Conclusion: Unlike previous works that mostly focused on just proposing a combination of DQN variants, this study combines DQN variants with actor-critic to benefit from both policy-based and value-based RL methods and overcome two main issues of both of them, slow learning and overestimation. Experimental results show that the proposed method can make a more accurate conversation with a user as a dialogue policy learner.
