Browsing by Author "Tanberk S."

Now showing 1 - 2 of 2
  • No Thumbnail Available
    Item
    Transformer Based Multimodal Summarization and Highlight Abstraction Approach for Texts and Speech Audios
    (Institute of Electrical and Electronics Engineers Inc., 2024) Altundogan T.G.; Karakose M.; Tanberk S.
    Multimodal summarization is a kind of summarization application in which its inputs and/or outputs can be in different data types like text, video, and audio. In this study, a new approach based on fine tuning of different pre-trained transformers was developed for abstractive and extractive summarization of audio and text data. In the proposed method, abstractive and extractive summaries of text data are provided only as text, while extractive summaries of audio data are presented as both text and audio data. Abstractive summaries of the audio data are presented as text only. Transformers with text2text input-output relationship were used in both extractive and abstractive summarization processes of the proposed method. For the training and inference processes of audio this type of data to be handled in transformers, an ASR step was followed before the summarization step. The experimental results obtained were given in detail and compared with similar approaches in the literature. As a result of the comparison, it was seen that the proposed method achieved better performance than similar prior approaches. © 2024 IEEE.
  • No Thumbnail Available
    Item
    Dynamic Fuzzy Cognitive Maps-Based Crowd Analysis Using Time Series Obtained From Video Processing
    (Institute of Electrical and Electronics Engineers Inc., 2025) Goktug Altundogan T.; Karakose M.; Yaman O.; Tanberk S.; Mert F.; Egemen Yilmaz A.
    Fuzzy cognitive maps treat the components of a problem or system expressed as fuzzy concepts and model the system with the relationships between these concepts. We predicted that FCM' s approach to calculating with these relationships could perform multivariate time-series forecasting with high performance. However, especially in real-world systems and related data sets, numerical values that clearly express the relationships between time series elements are not included, and this is a challenge. Another challenge that FCMs have for these problems is that real-world systems are highly dynamic structures and these relationships have variable properties in different situations. We evaluated these challenges as the main motivation factor and developed a GA-based method to determine system relationships for different states of the system. Then, in order to dynamically handle these relationships determined for different states on FCM, we took advantage of neural architectures that take the initial concept vectors as input and calculate the relationships between these concepts. In order to evaluate the performance of the time-series forecasting approach we developed, we performed time-series forecasting on two different scenarios using an artificially generated data set and a benchmark data set containing real-world data. In this way, we saw that the time series modeling performance of our proposed system is over 95%. FCMs perform the calculations they have made until the system becomes stable. This allows time series analyses to be performed not only depending on time but also depending on the steady state. Therefore, using this capability of the approach we developed to model time series formed from crowd analysis data obtained with video analytics is quite suitable in terms of providing the contribution points we present in this article. In this context, we integrated our FCM-based time series forecasting approach to two different video analytics scenarios by applying two different crowd analysis approaches we developed within the scope of this study. The success of our proposed method is over 95% both for performing for forecasting the time series obtained as a result of crowd analysis. Beside, crowd analysis approaches developed in this study have similar performance to state of the art approaches in the literature. © 2013 IEEE.