Abstract
Humanitarian crises are unpredictable and complex environments, in which access to basic services and infrastructures is not adequately available. Computing in a humanitarian crisis environment is different from any other environment. In humanitarian environments the accessibility to adequate computing and qualified human resources is usually limited. Hence, developing advanced computing technologies is a difficult task to accomplish in such environments.
Moreover, time and resources in those environments are also limited and devoted for lifesaving activities, which makes computing technologies among the lowest priorities for those who operate there. In humanitarian crises, interests and preferences of decision makers are driven by their original languages, cultures, education, religions, and political affiliations. Hence, decision making in such environments is usually hard and slow because it solely depends on human capacity in absence of proper computing techniques.
In this research we are interested in overcoming the above challenges by involving machines in humanitarian response. We aim to embed historical humanitarian data in a vector space so that machines can use this embedding in knowledge inference, question answering, and document classification. The goal of this research is to harness the power of artificial intelligence to provide answers through knowledge inference, which helps decision makers in speeding up humanitarian response, reducing its cost, saving lives, and easing human suffering.
This research is motivated by the researcher’s interest in saving time and resources for those who respond to humanitarian crises, which will eventually save lives and reduce suffering for those who are affected by humanitarian crises. The main contribution of this research is a semantic text classification model, through which humanitarian actors are able to take informed decisions, in response to their day-to-day challenges at low cost, resources, and time. The model has been developed using machine learning and natural language processing.
The above results can be achieved using several methods and techniques, among which is the use of natural language processing to clean and pre-process the historical humanitarian data, semantic sentence encoding to enrich our dataset by incorporating developmental aspects to humanitarian records, text augmentation to amplify the dataset, text embedding to capture the semantics of the humanitarian data, and text classification to extract and infer knowledge from the model.
The model has been evaluated quantitatively by measuring accuracy, precision, recall, and F1-score. It has been also qualitatively evaluated by performing four information retrieval experiments: analogy resolution, similarity detection, question answering, and document classification. In qualitative evaluation we conducted a ground truth check to assess the validity of the results provided by the model. Our model showed sound results in both quantitative and qualitative evaluations.
The potential outcomes of this research encompass three areas: Firstly, facilitating the reduction of human intervention in decision support in humanitarian crises, using an agile and low-cost solution to generate knowledge for humanitarian response; Secondly, enabling the reuse of the old humanitarian records to produce a humanitarian domain-specific embedding through which knowledge extraction can be facilitated, and; lastly, bridging the knowledge gap between humanitarian and development fields, allowing practitioners from both domains to be prepared for the aftermath of the crisis.