AI evaluation of emails and social media posts
One example is evaluating emails in terms of customer satisfaction. Nowadays customers can submit thousands of opinions on their experiences with products or services every single day through a wide range of electronic channels, via email, on rating websites or on social media platforms. Who wants to spend time and energy inputting and assessing this flood of data? These kinds of jobs are extremely tiring over time and almost impossible for a human to cope with.
Airlines, for instance, receive thousands of comments every day from their passengers, written in many different languages without a clear structure. These comments contain valuable information that says a great deal about the satisfaction – or dissatisfaction – with the company and its services. Analyzing this information manually is a long and laborious task. AI is already capable of evaluating this information and drawing conclusions from it. It understands incomplete or incorrect sentences and words and knows that “not sufficiently cooled beer” simply means “the beer was too warm.”
AI systematically forwards customer requests
This kind of information can be found in a wide range of documents, and in most cases lies unused in various databases. However, thanks to AI and text mining, this valuable data can be recorded in an automated process. AI is capable of categorizing topics and initiating actions, such as systematically forwarding a customer request to the right person. In an invoice, AI can understand was the invoice is for and allocates the invoices to the correct business transaction. The AI needs to understand the text in order to allocate it to a particular topic, something that – in the past – only humans have been capable of.
AI must learn and, using the known information, develop a model that can then allocate new documents to specific topics with a high level of precision. It sounds like a great deal of work, but that’s not the case. A conventional laptop can generate a model from 300,000 documents in a matter of minutes. Once the model has been generated, it only takes a couple of milliseconds to categorize a new document.
Speech and sound recognition with machine learning
Things start to get a little more difficult once we move into the realm of recognizing audio signals such as language and sounds. The first speech-recognition programs to enter the market ten years ago were relatively disappointing. They didn’t really help out too much. But now speech recognition has progressed significantly thanks to AI. Experts believe that a computer’s understanding of language is similar to that of a real person. Any users of Alexa, Siri or Google Assistance will know how far advance audio AI already is. Google Assistant is able to reserve a table for you in a restaurant and talk to the employee there.
Speech and sound recognition are based on machine learning, which is a part of AI. As in the case of text recognition, algorithms recognize audio patterns using existing data. But first the developer must feed the AI with relevant data and define the rules under which the data is analyzed. Essentially AI will find anomalies in audio data, even in the presence of background noise. The non-profit organization Rainforest Connection uses an AI-supported sound recognition system to track illegal logging in tropical rainforests. Solar-powered mobile phones record any continuous sounds and upload them to the Cloud, where AI can then determine any anomalies that might mean a truck is making its way through the forest or someone is using a chainsaw. A similar project has been implemented at the LHIND headquarters in Norderstedt. Sometimes the 30 elevators in the building will start making strange sounds, which indicates that a problem might be just around the corner. Taking pre-emptive action saves hugely on disruption and repair costs. But who among us would want to ride in an elevator all day and listen to the noises it makes? AI does – 24 hours a day, 7 days a week. As soon as a “strange” sound is heard – which the AI system has been taught to recognize – the AI raises the alarm.
Recognizing objects and people in video data
Machine learning is also used to analyze photo and video data. Here, AI experts frequently use deeply nested neural networks – or deep learning as explained above. To show how the process works, take the following example: A company wants to create a system that automatically recognizes the name of an object or a person in an image. In this example, the algorithm needs to differentiate between apples and pears. To do so, it has to be trained first. It is shown 200 images of various types of apples and pears. The algorithm learns by studying the images and internalizing the difference between an apple and a pear so that it can recognize in the future whether the image is showing an apple or a pear.
The AI is also able to analyze a continuous stream of video and recognize faces and persons independently. On screen the visitor watches the live images from a camera in which people, faces, handbags or similar items are marked. One essential aspect here is the live anonymization, in which the AI anonymizes the people in real time. The AI analyses the video frame by frame and can therefore follow a person or an item on its journey. This kind of video analysis can be used in security-related applications for example. One of the main advantages of this in-house solution is that the data is stored locally on the company’s computer and can be protected accordingly. AI-based automatic content extraction from image and video data raises the efficiency of content-related analysis.
AI solutions at a glance
- AI solutions at a glance
- Forwarding email requests to the right Person
- Cross-checking invoice information
- Categorizing customer feedback on social media channels or in emails
- Digitalizing printed invoices
- Recognizing objects in Images
- Analyzing audio data
- Manual visual inspection whether all products have been correctly sorted