Abstract
Manufacturing companies and product engineers struggle to develop effective
ways to detect and handle defective issues linked to their products due to the
lack of streamlined ways to go through the customer’s online reviews to locate
possible root causes associated with the defect information stated by the customers.
The lack of a practical and automated approach can lead to delays in
product recalls and recoveries. From the manufacturer’s standpoint, these risks
can derail their business.
This research explores deep learning techniques, namely, recurrent neural networks
(RNN), Convolution neural networks (CNN), attention-based topic mechanisms,
and natural language processing techniques to extract defect information
and possible root causes from the online reviews. This research proposes a deep
learning approach to give manufacturers possible ways to defect and flag product
defects. The RNN layer is implemented to model sentiment analysis in online
reviews. Then CNN layer is utilised to detect the kind of defect is mentioned in
each of the online reviews. Lastly, the result of the RNN and CNN layers are
fed into the attention-based topic model to extract a summary of keywords that
provide insight related to the defect information from the online reviews linked
to the products.