Research On Landmark-Oriented Multimodal Aspect-Opinion Summarization Mining

Accompanied by mobile terminal price decline and Wi-Fi device widely laid, the mobile Internet application has been experiencing an explosive growth. Based on the mobile terminal devices, users have generated a lot of landmark comment text and picture messages. The informative content of landmark provides excellent environment and urgent needs for research tasks of landmark opinion data mining.As the continuation work of development of the search and mining system on the basis of the Internet of things project in the laboratory, our goal is to get the most intuitive landmark aspect-opinion summary information from the large scale of web information. This research work can help users quickly fetch the key query information, make tourist destination decisions and plans, promote health tourism to flourish, promote information consumption and maintain economic growth.To better understand the city, we propose a novel framework termed multimodal aspect-opinion summarization (MAOS) to discover the aspect opinions about the popular scenic spots in Beijing. Firstly, we design the web crawler to obtain the reviews and travelogues from the travel websites. We have done the data collection and preprocessing work, selecting important informative texts and images through the entropy indicator and so on. Then, we begin the text mining work and generate the landmark textual aspect-opinion. We creatively promote the incremental learning method to identify the aspect-opinions from the selected sentences with the separation and cohesion indicator synchronously. Experiments show that the purity precision of the proposed incremental learning method has improved about19percent against the state-of-the-art approaches. Besides, images from travelogues will be clustered into several groups through spectral cluster algorithm, with the AP algorithm selecting some representative images from each cluster. Experiments show that the clustering and selected representative images are effective enough to express the scenic spots. The last but not the least, we get relevant and representative images to visualize the aspect opinions through textual combination method. Experiments show that the multimodal summary is readable, innovative, and information-rich.We have done extensive experiments on a real-world travel and review dataset to demonstrate the effectiveness of our proposed method. The multimodal summary is vivid, intuitively offering users key information about the landmarks and saving the search time. User study about the multimodal summarization including the text aspect-opinion mining and representative images get average high score, proving that users give positive altitude about the summary research work.