Our clients often ask us how our language technology is able to detect questions that are formulated differently, contain errors, or combine multiple questions in one text.
In order to answer these questions, we have written this article to provide a general overview of our technology. The descriptions are intentionally kept simple to ensure easy understanding. This article is primarily intended for service managers and business managers who want to automate their customer service.
What we do
We have been working in the field of NLU (Natural Language Understanding) / NLP (Natural Language Processing) for several years. Together with international research institutes, universities, as well as corporate partners, we have advanced important developments in the field of language processing.
We monitor developments in the research field of natural language proces’‘sing very closely and evaluate new developments worldwide in order to adopt new techniques quickly.
With all new developments in the field of language processing, we integrate these techniques based on the following three criteria:
The customer and customer service agent see their answers on the website or in the ticketing system without any delay. They can make adjustments to the question or search terms in real time.
All the knowledge required for the service is maintained in a database. All products continuously learn from each other and thus understand the customer’s intent with increasing precision.
All OMQ products can be used without any complex training. Once a list of frequently asked questions and answers has been added, the system can be used immediately.
Questions and answers
A software that automatically answers customer inquiries must be able to understand the customer’s intent and then display the right answer. The database contains a record of questions with the corresponding answers. The AI compares the texts (questions) with each other and determines their degree of similarity. If the degree of similarity is high enough, the answer can be displayed. A company’s database contains this record, for instance:
However, in most cases, customers will not ask the same question word-for-word. Although customer requests have the same intent, they are often formulated differently:
In order to be able to associate texts with each other, OMQ’s artificial intelligence understands that the phrase “Unfortunately, the temple was broken” has the same meaning as “How can I make a complaint about my glasses?”
OMQ’s artificial intelligence learns from customer inquiries. It analyzes words and how they relate to each other. Words with similar semantic relationships become associated. For example, let’s say the database contains the question “How can I reset my password?”, while the customer types “I forgot my login.” . In this case, the customer’s text is processed as follows:
Given the words’ semantic relatedness, OMQ’s artificial intelligence is able to associate “password” with “login” and “forgot” with “reset”. The words are mapped to proximate points in a multidimensional space based on their semantic relatedness. This space consists of thousands of dimensions. Words that can be associated with each other form clusters due to their proximity to each other. Each dimension represents a meaning and the words contained therein are closely associated with that meaning.
Word and question concepts
For example, for a software manufacturer in the multimedia sector, the following concepts can be relevant:
When texts are compared with each other, the AI does not use words but the previously created concepts to create a similarity measure. The questions in the database will in turn also be mapped as concepts of these word concepts. The database record “When exporting the created movie, an error message is displayed?” is formed from the concepts “export”, “movie”, and “error”.
The advantage of this technique is that the learned concepts can also be used by newly created questions. Questions thus learn from each other across levels.
Customer and agent feedback
In addition to analyzing semantic relatedness, OMQ’s artificial intelligence uses feedback from customers and customer service agents. The concepts are continuously modified and adapted based on this feedback.
As the word concepts change, so does the position of the individual questions. Questions that are similar move closer together, while questions that are different move away from each other. Through this continuous adaptation, questions that are not asked directly also benefit from user feedback. Thus, the entire knowledge base is constantly expanding.
In addition to the learning process, OMQ’s technology also uses linguistic information when analyzing texts. Dictionaries for various languages are used. For example, the software determines whether a word is relevant or which word family it belongs to. In addition, there are various concepts that describe fixed expressions, such as email addresses.
Since customers in the consumer sector sometimes have problems with spelling, OMQ’s artificial intelligence is also able to process misspelled words. This also applies to grammar and syntax. In support queries, customers often use punctuation marks incorrectly or not at all.
Oftentimes, customer inquiries are very long and contain a lot of irrelevant information. For instance, customers will write about how they have bought a product and are generally satisfied with it, and that they like a certain feature in particular, but get an error message with another feature.
OMQ’s artificial intelligence identifies the relevant passages as well as their intent. For this we use a technique that recognizes the relevant sections and creates an association for them. From this association, passages are formed, from which an intent is then recognized. Based on the intent, the AI searches the database for the correct answer.
The passages are segmented by topic. In order to resolve overlaps, we have developed a technique that determines the most likely segmentation based on the knowledge base. The identified sections are text parts, sentences, or multi-sentence descriptions of issues.
The extracted passages are then used in the second step to find an answer. OMQ Reply also uses these passages to create quotes from customer queries and then add an answer to them.
OMQ’s technology works in 32 different languages. The language of the entered text is detected and then selected for handling the customer’s query. So, for example, an English query can be entered on a German contact form that is linked with OMQ Contact. German answers are displayed for German queries, English answers are displayed for English queries, etc.
For language detection, we use a combination of an evaluation of rows of letters and words that occur frequently in a language. When syllables are recognized, a kind of fingerprint of a language is created. This is then compared with the text. For example, here are some common letter combinations for German and English:
The following word types and words are examples of common words that are often used in a language:
All our products use the OMQ engine. On our clients’ website, the help/FAQ page OMQ Help and contact form OMQ Contact are integrated. In call centers, OMQ Assist is integrated into the ticket/help desk. Emails are automatically answered by OMQ Reply.
All products access the same knowledge database and virtually use the same engine. There are only slight variations between products.
If you would like to learn more about us, our products, or our technology, please feel free to get in touch with us: Send us a message