Video Mining
With the widespread adoption of video sharing websites, scholars transfer their attention to digital videos mining. The video can be defined as one kind of content-based multimedia data which is typically analyzed from the viewpoint of specific video semantic annotation- text, audio and visual information. Zhang (2012) proposed one efficient video mining schema that combined object recognition, continuous speech recognition and video caption text recognition. He applied the dense sub graph finding approach to explore the semantic relationship between two neighboring words that only reserved the noun and verb words. Additionally, in terms of feature-based video mining approach, it can be categorized as video clustering mining, video classification mining and video association mining. The video clustering mining leverages the clustering algorithms such as k-means method on organizing the videos based on their homogeneous feature objects (Latecki & Wild, 2002). As for the video classification mining, it emphasized to dig out the implicit patterns among video objects like the semantic descriptions. In practice, Saravanan and Srinivasan (2010) also focused on the attribute extraction- fields of image processing, segmentation, edge detection, pattern recognition to design one efficient video framebased retrieval system. After grouping all the extracted features from videos, this structured data can be examined if there exists any associated patterns by the association rule (Xie & Chang et al., 2003). Apart from using low-level features with little meanings for naïve users, researchers (Zhu, Wu, A.K. & Wu, 2005) designed a knowledgebased video indexing and content management framework for sports domain specific videos. They took advantage of multilevel sequential association rule to explore the relationship between the audio and visual cues.
In terms of the interactive characteristic of videos, majority of the studies concentrate on analyzing the components of the video itself rather than the subjective facts provided from the viewers. However, the audience is the most significant key to determine the influential effect of the videos. Hence, instead of studying the video’s elements, the true experience of viewers is the main focus of this study, involving the classification of viewers’ preference and their attitudes toward this video.
Two Step Clustering Analysis
Customer segmentation is the primary marketing emphasis for effectively positioning the right role of portfolio strategies. Consider the scalability and complexity of data, researchers (Chiu, Fang, Chen, Wang & Jeris, 2001) demonstrated the design of two step clustering algorithm that performed well for mixed type attributes in large database environment. The fundamental procedures of this two-stage approach is to execute a pre-clustering step by the decrease in log-likelihood distance measure at first and then conduct a modified hierarchical agglomerative clustering algorithm to categorize the dense regions sequentially into homogenous clusters (Mooi & Sarstedt, 2011). In general, the number of clusters is automatically assessed by calculating measures of fit such as Akaike Information Criterion or Bayesian information criterion (Schwarz, 1978).
Literatures showed many successful applications of the two stage clustering method that contribute to the gain of competitive edge. In terms of customer segmentation in one Pakistan mobile telecommunication company, researchers (Salar, Moaz, Faryal, Ali, Aatif & Ahsan, 2013) adopted customers’ daily call and SMS usage as well as revenue generation data, discretized via binning method, to do the classification and uncover the usage behaviors.
Moreover, some scholars also took advantage of data mining technology to develop the new type of two-step clustering approach. Namver, Gholamian & KhakAbi (2010) leveraged RFM (Recency, Frequency, Monetary Value), demographic and customer lifetime value data to construct one new customer segmentation model. The mechanism they developed is to leverage k-means technique on the construction of intelligent customer segmentation based on the two-phase clustering schema. This study aimed at customer data in banking industry and grouped the existing customers according to their shared transactional behavior and characteristics. Through the analysis, this research help marketers establish better customer relationship management strategies, reduce the churn and find the good opportunities for up and cross selling.
In practice, the two step clustering algorithm is not merely useful for marketing use but also for the behavior prediction. The study of Higgs and Abbas (2013) revealed that each driver showed a unique distribution of behavior, but some of the behaviors
existed in more than one driver but at different frequencies through the two-stage clustering method. Regarding the adoption of this methodology in this research, it is definitely necessary for the process of variable transformation. In this study, the number
of segmentation, categorized by users’ subjective preferences and their social influencetotal upload views and subscription counts, is examined by means of the two-step clustering algorithm, proposed by Chiu et al. (2001).