Improving User Intent Detection in Urdu Web Queries with Capsule Net Architectures

Detecting the communicative intent behind user queries is critically required by search engines to understand a user’s goal and retrieve desired results. Due increased web searching in local languages, there an emerging need support language understanding for languages other than English. This article presents distinctive, capsule neural network architecture detection from Urdu, widely spoken South Asian language. The proposed two-tiered utilizes LSTM cells iterative routing mechanism between capsules effectively discriminate diversely expressed intents. Since no Urdu dataset available, benchmark intent-annotated of 11,751 was developed, incorporating 11 query domains annotated with Broder’s taxonomy (i.e., navigational, transactional informational intents). Through rigorous experimentation, model attained state art accuracy 91.12%, significantly improving upon several alternate classification techniques strong baselines. An error analysis revealed systematic patterns owing class imbalance large lexical variability queries.