Artificial intelligence (AI) has advanced tremendously, and Salesforce is taking it to the next level.
As with any technology, a tool is only as great as how one uses it. With Artificial Intelligence, the level of information that is returned can become overwhelming. Salesforce is attempting to improve the accuracy of machine learning to avoid information overload by automatically summarizing the text, tackling one of the most difficult aspects of AI, natural language processing (NLP).
Salesforce’s automatic summarization tool will not only resolve the issue of spending too much time sifting through e-mails, but it will allow sales and customer service teams better technology to allow them to spend more time with clients and customers. Their new algorithm introduces two new advances: “Contextual word generation model” and “A new way of training summarization models.”
The automatic text works in two ways:
- Extracting: The computer performs a “copy-and-paste method by extracting preexisting wording in the text. Similar to predictive words for iPhone messaging. While this is helpful, it is limited in terms of flexibility.
- Abstracting: The computer is able to generate a summary and introduce new works. The issue with this method is that the system does not have the ability to understand the original article.
The way that this plays together is through deep learning. The system has to analyze numerous examples of sentences and words. Through this method, the computer can adapt and learn how to interpret texts to then introduce it’s own suggested words. This process is done through reinforcement learning and scoring the metrics. Reinforcement learning is becoming more common for tasks that require generating language.
Meet ROUGE (Recall-Oriented Understudy for Gisting Evaluation. ROUGE is the metric system that is used. The algorithm is able to update itself by scoring against ROUGE. Salesforce tested out ROUGE and while the results may seem minor to outsiders looking in, their results were significant in the world of machine learning. Their model allows for fast analysis and accurate when it comes to a generated summary.
Salesforce is continuing to prioritize their testing, in hopes of rolling out this technology in the near future.
Read more here.