The capabilities of Artificial Intelligence (AI) and Machine Learning (ML) have successfully enabled them to enter into every possible industry. With the introduction of Large Language Models (LLMs) and Question answering systems in recent times, the AI community has advanced to a great extent. Efficiently retrieving responses from pre-computed databases containing question-answer pairings is a common step in the development of automated Question-answering (QA) systems.
There are two main QA paradigms: open-book and closed-book. The open-book paradigm, or Retrieve-and-read, is a two-step procedure in which pertinent material is obtained from a sizable document corpus, frequently the internet, and the solution is then taken out of the stuff that has been obtained by applying different models and methods. The closed-book method, on the other hand, is more recent and depends on skills learned in training as models using this paradigm, which are usually based on Seq2Seq models like T5, produce results without utilizing outside corpora.
Though closed-book techniques have shown remarkable results, they are too resource-intensive for many industrial applications and pose a significant risk to system performance. Database QA (DBQA) is another method that retrieves the response from a pre-generated database of question-answer pairs instead of depending on the information included in the parameters of models or sizable corpora.
A database of questions and answers, a retrieval model for querying the database, and a ranking model for choosing the best answer are the three main parts of these systems. DBQA techniques enable quick inference and the capacity to add new pairs without retraining models, thus introducing fresh information.
The lack of substantial training data is one of the main issues with DBQA techniques retrieval and ranking model development. Existing resources are scarce in terms of scope and content as a large number of them either need to improve in the quality of the annotation process or only concentrate on question-to-question similarity, thus ignoring replies.
To overcome these challenges, a team of researchers has proposed a dataset and models for question-answer database retrieval called QUADRo. It is a new, open-domain annotated resource that has been specifically made for training and assessing models. There are thirty related question-answer pairs for every one of the 15,211 input questions in the repository. This collection has a significant 443,000 annotated samples in total. A binary indicator indicating each pair’s importance in relation to the input query has been labeled.
The team has also undertaken a thorough experiment to assess the resource’s quality and characteristics in relation to several important QA system components. These elements consist of training methods, input model configuration, and relevancy of the answers. The experiments have demonstrated how well the suggested method works to retrieve pertinent responses by examining the behavior and performance of models trained on this dataset.
In conclusion, this research addresses the deficiency of training and testing data in automated quality assurance systems by introducing a useful resource and by carefully evaluating the resource’s attributes. A thorough grasp is aided by the emphasis on important elements like training tactics and answer relevancy.
Check out the Paper. All credit for this research goes to the researchers of this project. Also, don’t forget to join our 33k+ ML SubReddit, 41k+ Facebook Community, Discord Channel, and Email Newsletter, where we share the latest AI research news, cool AI projects, and more.
Tanya Malhotra is a final year undergrad from the University of Petroleum & Energy Studies, Dehradun, pursuing BTech in Computer Science Engineering with a specialization in Artificial Intelligence and Machine Learning.
She is a Data Science enthusiast with good analytical and critical thinking, along with an ardent interest in acquiring new skills, leading groups, and managing work in an organized manner.