Top 10 Conversational AI Research Papers that Techies Should Study

These conversational AI research papers are an asset for techies to explore more in the field of AI

Each year scientists from around the world publish thousands of research papers on AI but only a few of them reach wide audiences and make a global impact in the world. Artificial intelligence researchers are thriving in the global tech market with their immense technical know-how and inspirational strategies. They dig deep into AI models and make their followers understand the clear concept of the current and predictions happening in the tech community. It is a very time-consuming process to be a professional AI researcher with all the basic and in-depth knowledge about cutting-edge technologies like AI and technologies related to it. The tech community looks up to these global artificial intelligence researchers to be updated with the current information and to evaluate the pros and cons of the technologies like AI models. This article lists the top 10 conversational AI research papers that techies should study.

Transferable Multi-Domain State Generator for Task-Oriented Dialogue Systems, by Chien-Sheng Wu, Andrea Madotto, Ehsan Hosseini-Asl, Caiming Xiong, Richard Socher, Pascale Fung

The Hong Kong University of Science and Technology and Salesforce Research research team addresses the problem of over-dependence on domain ontology and lack of knowledge sharing across domains. In a practical scenario, many slots share all or some of their values among different domains (e.g., the area slot can exist in many domains like a restaurant, hotel, or taxi), and thus transferring knowledge across multiple domains is imperative for dialogue state tracking (DST) models.

Semi-Supervised Classification with Graph Convolutional Networks, Kipf, and Welling, ICLR 2017, cited by 7021

Discoveries of new drugs or efficient energy storage catalysts require modeling molecules as graphs. Graph convolutional networks brought the toolkit of deep learning into the graph domain, showing its superiority to hand-crafted heuristics that dominated the field before.

Do Neural Dialog Systems Use the Conversation History Effectively? An Empirical Study, by Chinnadhurai Sankar, Sandeep Subramanian, Chris Pal, Sarath Chandar, and Yoshua Bengio

The recently introduced generative models are good at generating fluent responses, but these responses tend to be boring and repetitive, which is often attributed to these models' poor understanding of dialog history. In this paper, the authors try to investigate empirically whether neural generative models use dialog history effectively.

Tabular Data: Deep Learning is Not All You Need (2021) – Ravid Shwartz-Ziv, Amitai Armon

To solve real-life data science problems, selecting the right model to use is crucial. This final paper selected by Max explores whether deep models should be recommended as an option for tabular data.

Persuasion for Good: Towards a Personalized Persuasive Dialogue System for Social Good, by Xuewei Wang, Weiyan Shi, Richard Kim, Yoojung Oh, Sijia Yang, Jingwen Zhang, and Zhou Yu

The paper builds on the Elaboration Likelihood Model, which argues that persuasive messages are more effective when they are tailored to people's worldviews. The authors recruited participants from Mechanical Turk, psychologically profiled them, and then asked them to role-play persuading each other to donate to the charity Save the Children.

Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks, Radford, et al., ICLR 2016, cited by 8681

GANs are machine learning models capable to generate new images of people, animals, or objects, and as such are responsible for the creativity of machines that gain popularity in photo-editing and designer apps. The proposed approach is now fundamental for all modern GAN models that generate new realistic images.

Competition-Level Code Generation with AlphaCode (2022) – Yujia Li et al.

Systems can help programmers become more productive. Asmita has selected this paper which addresses the problems with incorporating innovations in AI into these systems. AlphaCode is a system that creates solutions for problems that require deeper reasoning.

OpenDialKG: Explainable Conversational Reasoning with Attention-based Walks over Knowledge Graphs, by Seungwhan Moon, Pararth Shah, Anuj Kumar, and Rajen Subba

The Facebook Conversational AI team introduces a novel approach to creating natural, human-like responses with engaging, contextually diverse information about different entities and their attributes. First of all, they collect a new parallel corpus OpenDialKG where each mention of an entity in a conversation is manually linked with its corresponding ground-truth KG path. Then, the authors introduce a new model called DialKG Walker that can learn knowledge paths among entities mentioned in the conversation as well as reasons grounded on a large commonsense knowledge graph.

Human-level control through deep reinforcement learning, Mnih et al., Nature 2015, cited by 13615

The algorithms behind manufacturing, robotics, and logistics have moved from hard-coded rules to reinforcement learning models. DQN is one of the most popular deep reinforcement learning algorithms, which showed superior performance in various applications, without incorporating manually engineered strategies into itself.

A Commonsense Knowledge Enhanced Network with Retrospective Loss for Emotion Recognition in Spoken Dialog (2022) – Yunhe Xie et al.

There are limits to the model's reasoning in regard to the existing ERSD datasets. The final paper selected proposes a Commonsense Knowledge Enhanced Network with a backward-looking loss to perform dialog modeling, external knowledge integration, and historical state retrospect. The model used has been shown to outperform other models.