Correlation Analysis of Users Gaming Records and Review Classification
碩士 === 國立臺北科技大學 === 資訊工程系研究所 === 104 === In recent years， Gaming Industry has been flourishing. Every gaming social website keeps a considerable amount of gaming comments and users’ records. If the connection between them is found effectively， it can be used as a reinforcement on suggestion analysis...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | zh-TW |
| Published: |
2016
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/55du45 |
| Summary: | 碩士 === 國立臺北科技大學 === 資訊工程系研究所 === 104 === In recent years， Gaming Industry has been flourishing. Every gaming social website keeps a considerable amount of gaming comments and users’ records. If the connection between them is found effectively， it can be used as a reinforcement on suggestion analysis of comments.
Comparing with the traditional use of Wisdom of Crowds(the support of comments) as weight adjustments， in the beginning of new games’ launches， comments that don’t win supports due to not having enough users summit replies or lacking of exposure bring out distorted phenomenon of weight adjustments. However， by importing users’ gaming records， it can not only gain the information faster but show more realistic weights.
This article applies auto-classification as its base， respectively imports conventional and unconventional indices(for example， the number of users’ achievements， products， gaming hours and so on for conventional indices; levels， budges， friends， etc. for unconventional indices) from users’ gaming records as features for classifying， probes the impact each index posts on the result of auto-classification， tests whether the connection exists and if its influence is great， and tries to find the relations between indices.
The experiment employs the world-famous computer gaming social website Steam as a case for study. From the outcome of the experiment， we can see every index from gaming records has a connection between the accuracy of classified results. If we aim at certain indices， calculate their weights， and increase them on classifier when learning， this can elevate the accuracy of classification. By further decoding the meaning of relations and designing a way of calculating， this can strengthen classifier’s learning effect and promote auto-classification’s accuracy.
|
|---|