Abstract
In conventional data mining methods, the output is either a description of input data or a prediction of unseen data. But the real-world problems usually require interventions in order to alter the current data specifications towards a desirable goal. Actionable knowledge discovery is a field of study specifically developed for this matter. Existing methods rarely tackled the problem of extracting actionable knowledge from social networks. Moreover, due to the dependencies among the underlying network data, extracted actions should be evaluated since the changes suggested by the actions may not be described by the model constructed so far. This enforces the refinement of the model to preserve the quality of extracted actions. In this paper we propose a new method for action mining which incorporates an action evaluation process overcoming the mentioned problem while focusing specifically on social network data. Such data contains valuable information based on the links inside the network where a change in some feature values may result in a chain of changes in others due to the dependencies conveyed by the links in the network. We use a state-of-the-art structural feature extraction method to capture the information of the dependencies inside the network. Our proposed method iteratively updates structural features which are incorporated in the action extraction process. In this process, we thoroughly examine the effects of the application of actions by discovering the impact of possible changes in the network. We call this phenomenon “change propagation”. According to our experimentations, our method outperforms the state-of-the-art methods in terms of action effectiveness and reliability with comparable efficiency.
Highlights
Data mining provides an array of methods to extract knowledge from raw data
The first mentioned goal can be handled by extending the techniques to extract structural features, and for the second one, we propose a novel change-aware approach which incorporates the effect of the changes in different areas of the network in the action extraction process by evaluating the quality of the actions and rebuilding the model if needed
The action mining core and evaluation process are described under the topic of ‘‘change-aware action extraction and evaluation’’ where we dive into the details of extending the feature extraction method to track the changes and how the iterative evaluation process is carried out
Summary
Data mining provides an array of methods to extract knowledge from raw data. This knowledge is a great help to solve today’s world problems as the volume of digital data grows tirelessly. A variety of predictive or descriptive models can be learned from such data using mining methods, but in many cases, the need exceeds these outputs and a real-world problem cannot solely rely on the models derived from data. Actionable knowledge discovery (AKD) provides techniques and tools in order to facilitate automated decision making which is the area of expert systems. AKD exploits machine learning methods for discovering knowledge and patterns so that they are informative and/or predictive and applicable in the sense that actions fulfil business expectations.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have