Article Preview
TopIntroduction
Today chatbots became necessary and have an essential role in multiple areas to facilitate immediate and effective access to information and reduce the recurring call to humans. A chatbot is a conversational agent whose objective is to speak effectively and instantly to Internet users using interactive text or/and voice skills. Ideally, it should discuss with them, understand their questions, analyze their preferences, predict their intentions, and meet their needs. Whatever the request complexity, the chatbot should provide an answer not only in the theme but also in time.
Since their appearance, the number of chatbots has grown to include 80% of businesses in 2020 and automate more than 90% of banking interactions in (Daniel et al., 2016). However, despite their proliferation, the vocal skills they possess and the Artificial Intelligence they hold, an optimal human-chatbot conversation is not yet achieved and a compromise between an immediate and satisfactory response remains a challenge (Serban et al., 2017). The chatbots performance does not exceed 60% as more than 40% of the questions still unanswered (Qiu et al., 2017; Sinha, 2020). The provided responses may be either predefined so already exist in a knowledge base or intelligent so extracted using an inference model. In any case, they are generally inconsistent or meaningless (Qiu et al., 2017). For example, a study conducted on Apple's SIRI chatbot showed that only 8% of its answers were correct, 34% were false and 20% of the questions remained unanswered (Fadeni, 2019). Also, an instant response is still not guaranteed (Pereira & Diaz, 2018). Indeed, it is not yet obvious to the chatbot to semantically interpret the users' requests, to analyze their preferences and to predict their intentions while they have particularly heterogeneous profiles and very different contexts.
The role of the vast majority of existing chatbots is limited to understanding a short and simple textual message. They can only interpret well-defined texts whose responses are pre-existing, and their performance deteriorates considerably in the case of a complex query or a long dialogue (Serban et al., 2017). Many textual chatbots already exist in different areas such as MilaBot (Serban et al., 2017) et AliMe (Qiu et al.,2017) in the e-commerce field, MOOC Buddy (Iftene & Vanderdonckt, 2016) and CSIEC (Jia, 2009) in the e-learning field, MediBot (Srivastava & Singh, 2020) and Caro (Harilal et al.,2020) in the e-health field, and AVA (Yu et al.,2020) in the e-finance field. These conversational agents are absolutely useful, but today's application areas require much more complex data such as an image or a video in order to provide more context, details and information relating to their situations. Indeed, the audiovisual data represent a major source of information to deal with users' limited knowledge of the terminology of the field in which they need support. For example, in the e-commerce field, a user is not necessarily a computer scientist and so could mis-understand the technical the characteristics of the computer he is looking for.
In this case, he could simply send the chatbot an image or a video summarizing his needs. Such a search is therefore not possible with simple text messages since the preferences cannot be explicitly given by the user. In the medical field, which is considerably delicate, the chatbot should discuss rigorously with the patients, analyze their conditions and carefully answer their questions. To prevent the patient from transmitting false information, an image (exp. radiography or prescription) could be transmitted. This could help the chatbot make a personalized diagnosis of the patients’ state of health to help him decide about the care to consider. Another usefulness of visual chatbots arise when the user knows all these details but does not have time to write them and he prefers to send an image. The role of the chatbot is then to analyze this image and understand the user's preferences in order to meet his needs.