No abstract

Our aim in this paper is to detect bank clients involved in suspicious activities related to money laundering, using the graph of transactions of the bank. Although we have a labeled real dataset, our target is not only to obtain relevant results on it, but also on random graphs in which typical anomaly patterns have been injected. So, we want simultaneously adequacy to the real data and robustness. Our method is based on designing new features; the most important are those resulting from the reduced egonet, which is the subgraph that remains from an egonet after eliminating the nodes connected with a single edge to the center; another feature is built by appealing to random walks and serves as indicator of circular flows. Our features are added to usual egonet features and a general anomaly detection algorithm, in our case Isolation Forest, serves to detect the anomalies. Experiments on the real data and a comprehensive set of synthetic data show that our approach is adequate, robust and better than some previous methods.

Dialogue generation for open-domain conversations is a difficult and open problem that, so far, has not been able to approach human-level performance. Recently, a popular solution is to apply a sequence-to-sequence architecture, similar to the machine translation problem. These models try to map the input -given as the previous utterances, to the output -the next utterance. Unfortunately, they usually tend to repeat sentences, often preferring dull responses, that end the conversation abruptly. Therefore, Reinforcement Learning techniques have been combined with the standard sequence-to-sequence models in order to avoid their shortcomings. Our model applies a Policy Gradient method that maximizes the expected reward of generating the next utterance given a history of previous utterances. The results show an improvement in diversity up to 0.16 -almost 10x higher than the model without RL, while keeping the responses relevant to the input message.

A fairly novel area of research, at the conjunction between Artificial Intelligence (AI) and Human-Computer Interaction (HCI), resides in developing conversational agents as more users prefer this type of interaction to conventional interfaces. In this paper, we present an open-domain empathic chatbot, encompassing two of the biggest challenges of dialog systems: understanding emotions and offering appropriate responses. Although these tasks are trivial for a human, it is difficult to create a system that can recognize others' feelings in a discussion. The proposed model is developed based on the Generative Pre-Trained Transformer and it uses two datasets, PersonaChat and Empathetic Dialogues to achieve an empathic chatbot with a cordial personality. The measured performance -18.20 perplexity, 6.56 BLEU score, and 6.56 accuracy -comes close to the state-of-the-art models, while offering a further refined dialogue persona.

Goal-Oriented Chatbots in fields such as customer support, providing specific information or general help with bookings or reservations, suffer from low performance partly due to the difficulty of obtaining large domain-specific annotated datasets. Given that the problem is closely related to the domain of the conversational agent and that data belonging to a specific domain is difficult to annotate, there have been some attempts at surpassing these challenges such as unsupervised pre-training or transfer learning between different domains. A more thorough analysis of the transfer learning mechanism is justified by the significant boost of the results demonstrated in the results section. We describe extensive experiments using transfer learning and warm-starting techniques with improvements of more than 5% in relative percentage of success rate in the majority of cases, and up to 10x faster convergence as opposed to training the system without them.