Performance comparison of clustering algorithms are often done in terms of different confusion ma... more Performance comparison of clustering algorithms are often done in terms of different confusion matrix based scores obtained on test datasets when ground truth is available. However, a dataset comprises several instances having different difficulty levels. Therefore, it is more logical to compare effectiveness of clustering algorithms on individual instances instead of comparing scores obtained for the entire dataset. In this paper, an alternative approach is proposed for direct comparison of clustering algorithms in terms of individual instances within the dataset. A direct comparison matrix called \emph{Prayatul Matrix} is prepared, which accounts for comparative outcome of two clustering algorithms on different instances of a dataset. Five different performance measures are designed based on prayatul matrix. Theoretical analysis shows proposed measures satisfy five important properties such as scale invariance, data invariance, permutation invariance, Â monotonicity and continuity....
2021 5th Conference on Information and Communication Technology (CICT)
Tonic (Shadja) is a central idea in Indian Classical Music (ICM). It is the base pitch that an ar... more Tonic (Shadja) is a central idea in Indian Classical Music (ICM). It is the base pitch that an artist selects to create the melodies in a Raag composition, and it acts as a reference for the artist and for tuning all the accompanying instruments. Therefore, automatic identification of Tonic is a crucial task in the computational approaches of ICM, such as Raag Identification, melodic phrase analysis, etc. In this article, we propose a deep learning-based approach using frequency histogram peaks for Tonic identification. We demonstrate that the proposed approach combining deep learning and peaks of frequency histogram in most cases provides very high results with small mean absolute error and is easier to deploy in production systems than the existing multi-pitch-based approaches. Finally, we provide a systematic error overview of our approach, which provides further insight into the benefits and drawbacks of our methodology,
This dataset is a collection of mel-spectrogram features extracted from Indian regional music con... more This dataset is a collection of mel-spectrogram features extracted from Indian regional music containing the following languages:<br> Hindi, Gujarati, Marathi, Konkani, Bengali, Oriya, Kashmiri, Assamese, Nepali, Konyak, Manipuri, Khasi & Jaintia, Tamil, Malayalam, Punjabi, Telugu, Kannada. Five recordings are collected for each language for four artists (2Male + 2Female) each. 2 artists out of 4 for each language are old veteran performers, and the remaining 2 are contemporary performers. Overall, the dataset includes 17 languages, 68 artists (34 Males and 34 Females). There are 340 recordings in the dataset, with a total duration of 29.3 hrs. Mel-spectrogram is extracted from a 1-second segment with a 1/2 second sliding window for each song. Extracted mel-spectrogram for each segment is annotated with language, location, local_song_index, global_song_index, language_id, location_id, artist_id, gender_id. _______________________________________________________________________...
2021 12th International Conference on Computing Communication and Networking Technologies (ICCCNT), 2021
Generating the captions from the input images is a crucial problem, it includes both computer vis... more Generating the captions from the input images is a crucial problem, it includes both computer vision and natural language processing. In this paper, we introduced a hybrid feature and sequence extractor-based deep learning model that helps in generating captions for the input images. The proposed generative model is designed using a deep convolution neural network (VGG-19) for extracting the salient feature vectors from the images. While producing the informative captions from these feature vectors, the LSTM (Long Short Term Memory) network is utilized. The model is trained to expand the probability of the target description sentence delivered to the training image. In the analysis, the model is trained with very renowned data sets like FLICKR-8K, and the correctness of the model is calculated utilizing the BLEU score granting the range 0 to 100. We compare our model based on the BLEU score with four other models.
IEEE Transactions on Computational Social Systems, 2021
The flow of information through active users in online social networks (OSNs) plays a major role ... more The flow of information through active users in online social networks (OSNs) plays a major role in forming natural social groups, popularly known as communities. Although structural and topological aspects of the network had been central to most of the community detection approaches, incorporation of information flow for community detection has been an emerging topic in the recent past. Often, the flow of information is studied as a traceable process called information diffusion. The flow of information in the network affects various factors like temporal characteristics, network attributes, or social attributes. The information diffusion process helps to extract this information including where and when information is generated and in what fashion the dispersion occurs. Thus, it has the potential to aid the community detection process in social networks. In this article, the deployment of the information diffusion process for community detection has been studied extensively. The study is mainly focused on how information flow affects various network properties and social facets and explored the possibility of deployment for community detection. Various information diffusion models and community detection algorithms have been discussed in the context of network properties and social facets. Current challenges, future directions, and modalities for the deployment of information diffusion in community detection have been discussed. In addition, various widely used datasets, evaluation metrics as well as evaluation methods for evaluating community detection algorithms are also detailed.
2017 IEEE 4th International Conference on Soft Computing & Machine Intelligence (ISCMI)
Community detection problem has great importance for better understanding of the relationships am... more Community detection problem has great importance for better understanding of the relationships among the nodes as well as the overall network. In this paper, Atom Stabilization Algorithm (ASA) is considered for identifying communities. Modified Isolability is used as an objective function. Isolability measures the ability of group of nodes to isolate them from rest of the network. The results are compared with four other methods in terms of five quality and five accuracy metrics. The experimental results show the competency of proposed approach.
Malicious URL detection is an emerging research area due to continuous modernization of various s... more Malicious URL detection is an emerging research area due to continuous modernization of various systems, for instance, Edge Computing. In this article, we present a novel malicious URL detection technique, called deepBF (deep learning and Bloom Filter). deepBF is presented in two-fold. Firstly, we propose a learned Bloom Filter using 2-dimensional Bloom Filter. We experimentally decide the best non-cryptography string hash function. Then, we derive a modified non-cryptography string hash function from the selected hash function for deepBF by introducing biases in the hashing method and compared among the string hash functions. The modified string hash function is compared to other variants of diverse non-cryptography string hash functions. It is also compared with various filters, particularly, counting Bloom Filter, Kirsch et al., and Cuckoo Filter using various use cases. The use cases unearth weakness and strength of the filters. Secondly, we propose a malicious URL detection mec...
2021 International Conference on Intelligent Technologies (CONIT)
The community detection problem has attracted a huge number of researchers from the scientific co... more The community detection problem has attracted a huge number of researchers from the scientific community because it helps to understand the relationship between structural and functional properties of social networks. In this paper, we propose a novel community detection approach namely, Closeness Similarity driven Information Diffusion based community detection (CSID) which makes full utilization of local topology for similarity computation followed by information exchange and modularity maximization. Most of the existing algorithms so far are not suitable to quantify both the local topology information and exchange of information between the nodes. Extensive experiments are carried out to perform a comparative evaluation on several real-world datasets. The experimental results show that the proposed approach gives better results in comparison to most of the baseline algorithms.
Performance comparison of clustering algorithms are often done in terms of different confusion ma... more Performance comparison of clustering algorithms are often done in terms of different confusion matrix based scores obtained on test datasets when ground truth is available. However, a dataset comprises several instances having different difficulty levels. Therefore, it is more logical to compare effectiveness of clustering algorithms on individual instances instead of comparing scores obtained for the entire dataset. In this paper, an alternative approach is proposed for direct comparison of clustering algorithms in terms of individual instances within the dataset. A direct comparison matrix called \emph{Prayatul Matrix} is prepared, which accounts for comparative outcome of two clustering algorithms on different instances of a dataset. Five different performance measures are designed based on prayatul matrix. Theoretical analysis shows proposed measures satisfy five important properties such as scale invariance, data invariance, permutation invariance, Â monotonicity and continuity....
2021 5th Conference on Information and Communication Technology (CICT)
Tonic (Shadja) is a central idea in Indian Classical Music (ICM). It is the base pitch that an ar... more Tonic (Shadja) is a central idea in Indian Classical Music (ICM). It is the base pitch that an artist selects to create the melodies in a Raag composition, and it acts as a reference for the artist and for tuning all the accompanying instruments. Therefore, automatic identification of Tonic is a crucial task in the computational approaches of ICM, such as Raag Identification, melodic phrase analysis, etc. In this article, we propose a deep learning-based approach using frequency histogram peaks for Tonic identification. We demonstrate that the proposed approach combining deep learning and peaks of frequency histogram in most cases provides very high results with small mean absolute error and is easier to deploy in production systems than the existing multi-pitch-based approaches. Finally, we provide a systematic error overview of our approach, which provides further insight into the benefits and drawbacks of our methodology,
This dataset is a collection of mel-spectrogram features extracted from Indian regional music con... more This dataset is a collection of mel-spectrogram features extracted from Indian regional music containing the following languages:<br> Hindi, Gujarati, Marathi, Konkani, Bengali, Oriya, Kashmiri, Assamese, Nepali, Konyak, Manipuri, Khasi & Jaintia, Tamil, Malayalam, Punjabi, Telugu, Kannada. Five recordings are collected for each language for four artists (2Male + 2Female) each. 2 artists out of 4 for each language are old veteran performers, and the remaining 2 are contemporary performers. Overall, the dataset includes 17 languages, 68 artists (34 Males and 34 Females). There are 340 recordings in the dataset, with a total duration of 29.3 hrs. Mel-spectrogram is extracted from a 1-second segment with a 1/2 second sliding window for each song. Extracted mel-spectrogram for each segment is annotated with language, location, local_song_index, global_song_index, language_id, location_id, artist_id, gender_id. _______________________________________________________________________...
2021 12th International Conference on Computing Communication and Networking Technologies (ICCCNT), 2021
Generating the captions from the input images is a crucial problem, it includes both computer vis... more Generating the captions from the input images is a crucial problem, it includes both computer vision and natural language processing. In this paper, we introduced a hybrid feature and sequence extractor-based deep learning model that helps in generating captions for the input images. The proposed generative model is designed using a deep convolution neural network (VGG-19) for extracting the salient feature vectors from the images. While producing the informative captions from these feature vectors, the LSTM (Long Short Term Memory) network is utilized. The model is trained to expand the probability of the target description sentence delivered to the training image. In the analysis, the model is trained with very renowned data sets like FLICKR-8K, and the correctness of the model is calculated utilizing the BLEU score granting the range 0 to 100. We compare our model based on the BLEU score with four other models.
IEEE Transactions on Computational Social Systems, 2021
The flow of information through active users in online social networks (OSNs) plays a major role ... more The flow of information through active users in online social networks (OSNs) plays a major role in forming natural social groups, popularly known as communities. Although structural and topological aspects of the network had been central to most of the community detection approaches, incorporation of information flow for community detection has been an emerging topic in the recent past. Often, the flow of information is studied as a traceable process called information diffusion. The flow of information in the network affects various factors like temporal characteristics, network attributes, or social attributes. The information diffusion process helps to extract this information including where and when information is generated and in what fashion the dispersion occurs. Thus, it has the potential to aid the community detection process in social networks. In this article, the deployment of the information diffusion process for community detection has been studied extensively. The study is mainly focused on how information flow affects various network properties and social facets and explored the possibility of deployment for community detection. Various information diffusion models and community detection algorithms have been discussed in the context of network properties and social facets. Current challenges, future directions, and modalities for the deployment of information diffusion in community detection have been discussed. In addition, various widely used datasets, evaluation metrics as well as evaluation methods for evaluating community detection algorithms are also detailed.
2017 IEEE 4th International Conference on Soft Computing & Machine Intelligence (ISCMI)
Community detection problem has great importance for better understanding of the relationships am... more Community detection problem has great importance for better understanding of the relationships among the nodes as well as the overall network. In this paper, Atom Stabilization Algorithm (ASA) is considered for identifying communities. Modified Isolability is used as an objective function. Isolability measures the ability of group of nodes to isolate them from rest of the network. The results are compared with four other methods in terms of five quality and five accuracy metrics. The experimental results show the competency of proposed approach.
Malicious URL detection is an emerging research area due to continuous modernization of various s... more Malicious URL detection is an emerging research area due to continuous modernization of various systems, for instance, Edge Computing. In this article, we present a novel malicious URL detection technique, called deepBF (deep learning and Bloom Filter). deepBF is presented in two-fold. Firstly, we propose a learned Bloom Filter using 2-dimensional Bloom Filter. We experimentally decide the best non-cryptography string hash function. Then, we derive a modified non-cryptography string hash function from the selected hash function for deepBF by introducing biases in the hashing method and compared among the string hash functions. The modified string hash function is compared to other variants of diverse non-cryptography string hash functions. It is also compared with various filters, particularly, counting Bloom Filter, Kirsch et al., and Cuckoo Filter using various use cases. The use cases unearth weakness and strength of the filters. Secondly, we propose a malicious URL detection mec...
2021 International Conference on Intelligent Technologies (CONIT)
The community detection problem has attracted a huge number of researchers from the scientific co... more The community detection problem has attracted a huge number of researchers from the scientific community because it helps to understand the relationship between structural and functional properties of social networks. In this paper, we propose a novel community detection approach namely, Closeness Similarity driven Information Diffusion based community detection (CSID) which makes full utilization of local topology for similarity computation followed by information exchange and modularity maximization. Most of the existing algorithms so far are not suitable to quantify both the local topology information and exchange of information between the nodes. Extensive experiments are carried out to perform a comparative evaluation on several real-world datasets. The experimental results show that the proposed approach gives better results in comparison to most of the baseline algorithms.
Uploads
Papers by Anupam Biswas