-
GAE-ISumm: Unsupervised Graph-Based Summarization of Indian Languages
Authors:
Lakshmi Sireesha Vakada,
Anudeep Ch,
Mounika Marreddy,
Subba Reddy Oota,
Radhika Mamidi
Abstract:
Document summarization aims to create a precise and coherent summary of a text document. Many deep learning summarization models are developed mainly for English, often requiring a large training corpus and efficient pre-trained language models and tools. However, English summarization models for low-resource Indian languages are often limited by rich morphological variation, syntax, and semantic…
▽ More
Document summarization aims to create a precise and coherent summary of a text document. Many deep learning summarization models are developed mainly for English, often requiring a large training corpus and efficient pre-trained language models and tools. However, English summarization models for low-resource Indian languages are often limited by rich morphological variation, syntax, and semantic differences. In this paper, we propose GAE-ISumm, an unsupervised Indic summarization model that extracts summaries from text documents. In particular, our proposed model, GAE-ISumm uses Graph Autoencoder (GAE) to learn text representations and a document summary jointly. We also provide a manually-annotated Telugu summarization dataset TELSUM, to experiment with our model GAE-ISumm. Further, we experiment with the most publicly available Indian language summarization datasets to investigate the effectiveness of GAE-ISumm on other Indian languages. Our experiments of GAE-ISumm in seven languages make the following observations: (i) it is competitive or better than state-of-the-art results on all datasets, (ii) it reports benchmark results on TELSUM, and (iii) the inclusion of positional and cluster information in the proposed model improved the performance of summaries.
△ Less
Submitted 25 December, 2022;
originally announced December 2022.
-
Quantum Inspired High Dimensional Conceptual Space as KID Model for Elderly Assistance
Authors:
Ishwarya M S,
Aswani Kumar Ch
Abstract:
In this paper, we propose a cognitive system that acquires knowledge on elderly daily activities to ensure their wellness in a smart home using a Knowledge-Information-Data (KID) model. The novel cognitive framework called high dimensional conceptual space is proposed and used as KID model. This KID model is built using geometrical framework of conceptual spaces and formal concept analysis (FCA) t…
▽ More
In this paper, we propose a cognitive system that acquires knowledge on elderly daily activities to ensure their wellness in a smart home using a Knowledge-Information-Data (KID) model. The novel cognitive framework called high dimensional conceptual space is proposed and used as KID model. This KID model is built using geometrical framework of conceptual spaces and formal concept analysis (FCA) to overcome imprecise concept notation of conceptual space with the help of topology based FCA. By doing so, conceptual space can be represented using Hilbert space. This high dimensional conceptual space is quantum inspired in terms of its concept representation. The knowledge learnt by the KID model recognizes the daily activities of the elderly. Consequently, the model identifies the scenario on which the wellness of the elderly has to be ensured.
△ Less
Submitted 19 November, 2018;
originally announced November 2018.
-
Graphene platelets reinforced aluminum matrix composite with enhanced strength by hot accumulative roll bonding
Authors:
Jitendar K Tiwari,
Ajay Mandal,
N Sathish,
Venkat A N Ch,
A K Srivastava
Abstract:
Accumulative Roll Bonding(ARB) process was used to develop few-layer graphene nano platelets reinforced aluminum matrix (GNPs/Al) composite in the form of sheets. Annealed Al sheets were ARB processed up to 6 pass along with coating of graphene between stacked sheets in first and second pass. Another set was prepared with same process parameter in the absence of graphene coating. Properties of the…
▽ More
Accumulative Roll Bonding(ARB) process was used to develop few-layer graphene nano platelets reinforced aluminum matrix (GNPs/Al) composite in the form of sheets. Annealed Al sheets were ARB processed up to 6 pass along with coating of graphene between stacked sheets in first and second pass. Another set was prepared with same process parameter in the absence of graphene coating. Properties of these two set of samples were analyzed on the basis of film theory of ARB process which enables micro-level mixing of material on stacked layer interface. Hence the problem of graphene agglomeration was overcome through our process. The Raman spectra at the cross-section was taken which not only show the strong interaction of GNPs with Al matrix due to increased D and D' band but also the graphene quality enhancement on the basis of single symmetric 2D band. Samples were then subjected to universal testing machine (UTM) and Vickers microhardness tester. Results showed up to ~73% increment in yield strength and ~27% increment in hardness of GNPs/Al matrix composite. Fracture surface of tensile specimen was further examined under SEM to understand the fracture mechanics at higher passes, which elucidate elongation variation and delamination behavior of stacked sheets. Deep elongated dimples with the smoothed surface in GNPs/Al composite are the cause of reduced elongation. But the final composite was having improved strength with appreciable ductility. The improvement was further justified by increment in dislocations, calculated using Williamson-Hall plot on X-ray diffraction (XRD) data. All results signify the effectiveness of the proposed technique for the development of GNPs/Al composite.
△ Less
Submitted 30 June, 2018;
originally announced July 2018.
