Page 177 - The-5th-MCAIT2021-eProceeding

Fig. 1. Newspaper summary slots example
The downside of having predefined or arbitrary length is there will be cases where the usage of the summary
is for a specific space/area. Magazine or newspaper editors tend to require summary that will fit into a certain
slot in a cover. The current state-of-the-art as discussed above do not address output area based summary.
Neither Saito et al., 2020 nor any of the other previous works are able to generate summary for the
aforementioned scenario.

3. Methodology
Key features will be taken from the analysed techniques and will be used to develop an algorithm that will
perform summarization based on length constraint. Finally, an intensive evaluation of the quality and relevance
of the summary will be done in order to verify the enhancements. It is anticipated that by the end of the research,
the new model will be more enhanced than the current length control techniques mentioned in this paper.

3.1. Output area based abstractive text summarization (Proposed)

Given the source text and an image with a portion where the summary will be portrayed (such as in
magazine/newspaper story covers), using image processing the area of the summary output portion is calculated
to determine the suitable summary length that will fit the portion. Next, the determined suitable length is
embedded into the encoder-decoder model sequence to sequence technique to perform the summarization.

Fig. 2. Proposed model

The idea is to obtain a summary with a desirable length without having to predefine a specific length but
rather produce the summary through analyses of where the summary will be outputted.

E- Proceedings of The 5th International Multi-Conference on Artificial Intelligence Technology (MCAIT 2021) [164]
Artificial Intelligence in the 4th Industrial Revolution