A new hybrid multi-start tabu search for finding hidden purchase decision strategies in WWW based on eye-movements

• The proposed HMTS algorithm can find hidden decision strategies by clustering the eye-movement sequences from users.
• HMTS uses adaptive memory programming and incorporates multi-start and local search strategies for global optimization.
• HMTS outperforms standard TS, GA, PSO, and K-means algorithms on both empirical and synthetic eye-movement datasets.
• The scalability and robustness analyses for HMTS have been conducted through a series of statistical tests.

It is known that the decision strategy performed by a subject is implicit in his/her external behaviors. Eye movement is one of the observable external behaviors when humans are performing decision activities. Due to the dramatic increase of e-commerce volume on WWW, it is beneficial for the companies to know where the customers focus their attention on the webpage in deciding to make a purchase. This study proposes a new hybrid multi-start tabu search (HMTS) algorithm for finding the hidden decision strategies by clustering the eye-movement data obtained during the decision activities. The HMTS uses adaptive memory and employs both multi-start and local search strategies. An empirical dataset containing 294 eye-fixation sequences and a synthetic dataset consisting of 360 sequences were experimented with. We conduct the Sign test and the result shows that the proposed HMTS method significantly outperforms its variants which implement just one strategy, and the HMTS algorithm shows an improvement over genetic algorithm, particle swarm optimization, and K-means, with a level of significance α = 0.01. The scalability and robustness of the HMTS is validated through a series of statistical tests.

Figure optionsDownload as PowerPoint slide