In our final article in the “Intelligent search slot” series, we will focus on the part that is visible and tangible for every user first and foremost: the auto-suggest (or autocomplete called). In other words, the feature that we have all come to know and appreciate from Google. In dozens of implementations, we at moresophy have repeatedly found that this function is critical for the acceptance of a search application. Not only because users are now so conditioned to it thanks to Google that they simply expect to receive meaningful suggestions, but because it is a highly effective tool. An intelligent suggestion mechanism not only leads to success faster, it can even create a dialog with the user.

How Does Google Autocomplete Work?

Let’s first take a look at Google – today’s benchmark par excellence. In the example shown, Google returns meaningful completions from very different domains when you enter “aut”. It is important to understand where these suggestions come from and what data Google uses. The search engine giant has a unique and gigantic pool of data at its disposal. First and foremost, Google’s autocomplete is fed by the billions of search queries that users make every day. Google can therefore measure user demand and interest very precisely. In conjunction with the regional context (IP of the user), such trends can then be filtered further. Drive-in cinemas are experiencing a revival in times of the coronavirus pandemic. And since we carry out our search query from Munich, we are offered regional drive-in cinemas. As we do not use a personalized search and Google is therefore not likely to know (if you take data protection seriously) what the individual user is personally interested in, Google also offers further queries from completely different domains, e.g. “autism”.

If the user now takes up the last suggestion, for example, Google learns that they are looking for information about autism rather than cars and begins to make specific offers in this area : e.g. “autism test”, “autism symptoms” or “autismus bayern”. This creates an interactive dialog in which the user is guided by the suggestions to specify their search interest in such a way that Google ultimately provides them with exactly the results they need. As amazingly simple as this looks, it is difficult for any other search engine operator – whether in e-commerce, customer service or media portals – to offer a comparable experience. This is because nobody has anywhere near as many search interactions and users to offer good quality via this mechanism.

How Can I Create a Convincing Autocomplete with Less Data?

Providing a good autocomplete requires intuition, experience and, first and foremost, a professional concept. Because not all searches are the same. The underlying algorithm should be tailored to the purpose of the search. For the sale of products in an online store, the suggestion mechanism should work differently than when answering inquiries in customer service or for a general information search in a specialist portal. In CONTEXTSUITE, we use a combination of different procedures and rules to quickly bring the user to the target or the provider to the conclusion. A good understanding of the searched content plays a decisive role here. This is because Google’s approach is not applicable due to a lack of statistics – especially not right from the start. A good autocomplete therefore uses a combination of content, user queries and editorial guidelines. In other words: relevant suggestions are those that

1. Lead to good results (content)
2. I know are frequently searched for (user queries)
3. I want to promote as a provider (editorial)

When evaluating the content, the algorithm naturally benefits decisively from the intelligent pre-processing processes that we described in Part 1 and Part 2 of this article series. Taking the content as the leading data source for the autocomplete has a very decisive advantage: it ensures that only suggestions are made that lead to good search results. This is also anything but a matter of course. It is only trivial for Google because the search engine has such a comprehensive amount of information in its index that it can deliver something useful for almost any query.

The example shown is from customer service. Autocomplete is optimized to “anticipate” a user’s possible questions and quickly provide the right answers in the form of suitable text excerpts. In e-commerce, on the other hand, it makes sense to take sales targets into account when generating suggestions and, for example, to give a higher weighting to current offers or products with a very high margin.

With the SMART SEARCH application of our CONTEXTSUITE, the implementation of such intelligent search solutions is fast and standardized. Thanks to a combination of semantic processes and deep learning algorithms, even the adaptation to new domains or company specifics is carried out without a great deal of manual effort. CONTEXTSUITE also supports comprehensive cognitive analyses of German and many other languages.