What Google Assistant,Siri, Alexa, Cortana have in common? They tell jokes about different cleverness,most of which are the work of the writing teams work behind the scenes. They are interesting, but preliminary research suggests they also play a role, so that the interaction assistant.
Of course, there is always room for improvement. In pursuit of the assistant the ability to cut jokes to the user’s personal tastes, Amazon-the researchers investigated a joke of the selection method, click on one of the basic natural language processing mode or the mechanical learning model. They say that when tested, the production data, these two methods of”positive”impact of user satisfaction and possible improvements of joke-telling.
Training mode needs a wide range of additional description of the data set, where the team compiled a record set of the voice assistant user’s reaction to jokes. Two implicit feedback strategy is employed, one in which a joke is labeled as”positive”(i.e., funny)if the user requires a new joke within five minutes of the hearing and the second, marked positive all the jokes please, followed by New within 1 to 25 hours.
Comparison of different labeling techniques, the team conducted an A/B test in a production setting, another comparison relates to the historical data and the selected standard strategies. Joke data set contains thousands of unique jokes across categories(for example, sci-fi and sports)and types(puns, limericks, etc.) is used to verify each model, use data from approximately 80,000 English-speaking”customer”in total one(probably be Alexa users, although the researchers are unclear to say so).
The results show that the proposed natural language processing model consistently outperforms the rule-based method for both marking strategies. They noted that the machine learning mode and the implementation in terms of accuracy and performance, but this building—which is far greater than the natural language model processing–will make it difficult to extend to new countries and languages.
Researchers leave to future work to compare other methods and to develop a method that can be easily extended to new languages.
A great sense of humor to enhance the assistant’s use in those who have not climbed on the bandwagon. Estimates from ad in August pegged the number of monthly users of the voice assistant in the approximately 1 billion 1 thousand 2 million, from 102 million in 2018, but a separate survey and the PricewaterhouseCoopers report found that the poor understand why the AI assistant has the ability to do so, the widespread lack of trust may hinder the sector growth.