This thesis is for structural analysis for natural language interface using H-Frame between smart TV and user. We can easily observe the omissions of core information and pronouns while conversations. Therefore it is not an easy task to process and un...
This thesis is for structural analysis for natural language interface using H-Frame between smart TV and user. We can easily observe the omissions of core information and pronouns while conversations. Therefore it is not an easy task to process and understand the utterance with only information contained in it. In order to complete the sentence that has insufficient information, our analysis should find the missing information through previous conversations and we need a way to manage entire conversations to build up sentence structures. To do this, we introduce a frame management structure called H-Frame, which contains the core information. With H-Frame, user’s utterance can be analyzed. And information retrieval and TV control can be processed. In particular, we show in this thesis that our approach can produce efficient and accurate results by using 2-gram methodology to recognize program titles in user’s discourses. In addition to this, we can also improve our results and control the smart TV by utilizing time information of user’s utterance.
The contents and results of this thesis can be summarized as follows: It is required to process what a user requests through analyzing utterance between the smart TV and the user. In order to do this, we need to analyze structures and semantics of the utterance. However, the omissions of words and pronouns in conversations happen. Therefore, it is required to find the missing information to complete the syntax. To address this, we proposed H-Frame methodology which manages core information, where H-Frame is used to fill up missing information for complete sentence structure. Our proposed approach can help to get accurate results because we can extend sentence structures by filling up the omitted information from H-Frame.
In this area, we can take advantage of the technique of H-Frame showing that 97% of incomplete sentences can be recovered to complete ones. In addition, two structures in H-Frame were used for usage extensions.: One to manage program being broadcast and the other to manage conversation. Such internal structures can help to use less memory and better accessibility of information than existing stack-based systems.
It is common to abbreviate the title of a program or to use only part of the title in user’s utterances. In order to recognize such utterances, two approaches were used. In the first approach, we use word-based 2-gram search methodology if an utterance is a part of a program title. In this approach, there is one exception if more than three syllables should be processed. In this case, we use 1-gram search methodology. In the second approach, we developed four combination rules for searches if abbreviation forms are used. We can detect abbreviated titles up to 97% in title database containing titles from 2006 to 2011.
We defined 23 semantic features extracted by analyzing the meanings of words used in sentences, which are necessary for this thesis. We also found 63 sentence structures in our analysis. In order to provide more accurate results, we use context awareness. Such context awareness utilizes time information when a user started talking, leading to meaningful results. In addition, it prevents from doing meaningless actions when a user controls a TV.
In this thesis, we showed that our approach, structure analysis natural language interface using H-Frame between users and smart TV, is more efficient than a stack-based system. If we are able to use EPG information which is not dealt in this thesis, we can build more effective system and expect more accurate results.
，韩语论文网站，韩语论文 |
