<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by searchhub.io Dev Blog on Medium]]></title>
        <description><![CDATA[Stories by searchhub.io Dev Blog on Medium]]></description>
        <link>https://medium.com/@searchhub.io?source=rss-de55c8251d7d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*w6lx-7Y_RybDm0WHOqxhJA.png</url>
            <title>Stories by searchhub.io Dev Blog on Medium</title>
            <link>https://medium.com/@searchhub.io?source=rss-de55c8251d7d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 15:10:42 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@searchhub.io/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[A closer look into the spell correction problem — Part 3 — the bells and whistles]]></title>
            <link>https://medium.com/@searchhub.io/a-closer-look-into-the-spell-correction-problem-part-3-the-bells-and-whistles-19697a34011b?source=rss-de55c8251d7d------2</link>
            <guid isPermaLink="false">https://medium.com/p/19697a34011b</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[search]]></category>
            <category><![CDATA[information-retrieval]]></category>
            <category><![CDATA[search-engines]]></category>
            <dc:creator><![CDATA[searchhub.io Dev Blog]]></dc:creator>
            <pubDate>Mon, 25 Sep 2017 11:40:39 GMT</pubDate>
            <atom:updated>2017-09-28T09:56:39.503Z</atom:updated>
            <content:encoded><![CDATA[<p>At <a href="https://www.searchhub.io/">searchhub.io</a> query cleansing of human input (user query) is the first strategy we apply to each and every search query we receive. In Part 1 and 2 we already discussed a little about the challenges of spell correction at scale and independency of language.</p><p>However at search|hub we strive to help software systems to understand humans. Therefore we not only have to cater for typos like “scatebord -&gt; skateboard”. There are a lot more reasons why a search engine might not understand or even worse misunderstand a user query.</p><p><a href="https://www.youtube.com/watch?v=cbtf1oyNg-8">https://www.youtube.com/watch?v=cbtf1oyNg-8</a></p><h4>1.Word Segmentation &amp; Word Decomposition:</h4><p>Since most search engines are still based on a token representation of words we first have to identify the words in a user query. This might seem easy and obvious but in quite some cases it’s not. And I’m not only talking about cases that occur in some languages which are traditionally written without inter-word spaces, like <a href="https://en.wikipedia.org/wiki/Chinese_language">Chinese</a> and <a href="https://en.wikipedia.org/wiki/Japanese_language">Japanese</a> or cases where user queries are produced by some sort of speech recognition system.</p><p>Let’s take a real-world example: “<em>Damenmotorradlederhandschuh</em>”. Now you might think WTF is this. This is a german compound word which is built by gluing together the following words translated into english. “women+bicycle+leather+gloves”</p><p><strong>The dictionary approach:</strong></p><p>The traditional approach tackling such a query would be to use a decomposition dictionary that scans through the query and tries to break the query as soon as it finds a sub-word from the dictionary. So let’s do that “<em>Damenmotorradlederhandschuh</em> -&gt; <em>Damen motor rad leder hand schuh</em>” Again for the ones that do not speak german: “women+engine+wheel+leather+hand+shoes”</p><p>Oh, wait what the heck happened here: by splitting the words we changed their meaning! Imagine the search result for such a query.</p><p>And what happens if I misspell the query?</p><p>“<em>Damenmotoradlederhantschuh</em> -&gt; <em>Damen motorad leder hantschuh</em>” So in this query the user made two simple errors and not even the mighty google is able to guess what the user was looking for.</p><blockquote>Word Segmentation &amp; Word Decomposition are vital parts of the query understanding process and you can’t fix this part at scale by manual mapping through dictionaries and ambiguity handling.</blockquote><h4>2.Primary Word Detection:</h4><p>Once you segmented / decomposed the query into words you’ll soon realize that now there is another query dimension you have to take care of. <em>The sequence of words</em>. There are several cases where the order or sequence of words inside a user query might change the meaning of the query or at least change the stemming approach.</p><p>Let’s directly jump in another example for this: Imagine the above user query “<em>Damenmotorradlederhandschuhe</em>” and a couple of other queries which represent the same intent/meaning -&gt; “<em>motorrad leder handschuhe damen, leder motorradhandschuh für damen</em>”.</p><p>In this example, the order or sequence of words is pretty much independent of its meaning or intent. However as soon as you want to introduce stemming you better make sure that you only stem the “primary word(s)” in this case “<em>handschuh(e)</em>”.</p><p>But not every user query that includes the same words represents the same meaning or intent. The query “<em>Armbanduhr</em>” aka wristwatch vs. “<em>Uhrarmband</em>” aka watch bracelet is a perfect example for this. Both queries segmented or decomposed consist of exactly the same words but describe two different things. To solve this problem we first have to identify those user queries and then find the primary word in order to understand its meaning or intent.</p><h4>3.Under- &amp; Overstemming:</h4><p>Grammatically correct stemming can be very tedious. Applying traditional stemmers like <a href="https://de.wikipedia.org/wiki/Porter-Stemmer-Algorithmus">Porter</a> or <a href="http://snowball.tartarus.org/">Snowball</a> usually leads to a lot of overstemming or understemming — especially with short words which represent the majority of the query corpus.</p><p>Again let’s take a real-world example: “<em>babybetten -&gt; babybetten</em>” and “<em>vans -&gt; van</em>” and <em>iphone5s -&gt; iphone5</em>. In the first example, the porter stemmer was unable to stem babybetten to its root babybett while in the second example the brand name vans was reduced to van which in this case changes its meaning.</p><p>But in order to retrieve relevant and meaningful search results the search engine needs to understand the meaning of the query. While singular and plural forms normally represent the same meaning / intent this might not be the case for automatically stemmed words.</p><h3>search|hub does all of this automatically</h3><p>When we build search|hub we solved all of these areas by combining domain knowledge, smart algorithms and machine learning models fueled by user data. We strongly believe that all of this is key to make search engines understand humans.</p><p>SEARCH IS THE PLACE WHERE THE USER IS TELLING YOU WHAT HE WANTS. IF YOUR SEARCH ENGINE SPEAKS THE SAME LANGUAGE AS YOUR USERS SEARCH BECOMES A CONVERSATION. SEARCH|HUB HAS SPECIFICALLY BEEN DESIGNED TO HELP YOUR EXISTING SEARCH ENGINE TO UNDERSTAND HUMANS AND DRIVE THESE CONVERSATIONS.</p><h3>We are hiring</h3><p>If you’re excited about advancing our search|hub API and strive to enable companies to create meaningful search experiences, join us! We are actively hiring for Data Scientists to work on next-generation API &amp; SEARCH technology.</p><p><a href="http://www.searchhub.io/">www.searchhub.io</a> proudly built by <a href="http://www.commerce-experts.com/">www.commerce-experts.com</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=19697a34011b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A closer look into the spell correction problem — Part 2 — introducing preDict]]></title>
            <link>https://medium.com/@searchhub.io/a-closer-look-into-the-spell-correction-problem-part-2-introducing-predict-8993ecab7226?source=rss-de55c8251d7d------2</link>
            <guid isPermaLink="false">https://medium.com/p/8993ecab7226</guid>
            <category><![CDATA[search]]></category>
            <category><![CDATA[search-engines]]></category>
            <category><![CDATA[information-retrieval]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[searchhub.io Dev Blog]]></dc:creator>
            <pubDate>Tue, 22 Aug 2017 14:50:34 GMT</pubDate>
            <atom:updated>2017-09-25T08:18:43.910Z</atom:updated>
            <content:encoded><![CDATA[<p>At <a href="https://www.searchhub.io/">searchhub.io</a> we need to spell correct millions of unique queries per minute. This sounds pretty straight forward but is actually a very hard problem to solve. There are several ways to approach this challenge and therefore we wanted to share our experience during the last 18 months.</p><h3>1.The Data-driven deep-learning way</h3><p>Building a spell checker using deep learning sounds like a great idea. After the recent hype of deep learning, we decided to give it a try and see how far we can get. In order to do so we built our own character-based model for spell checking. We used our query test data set which essentially consists of spelling errors and their representative correction. The dataset is split into a training set and a dev (validation) set in 90/10 proportion. Then an LSTM sequence-to-sequence model is trained to predict a corrected query based on the query with typos. If you are interested in more details around the model you should read Tal Weiss’s <a href="https://medium.com/@majortal/deep-spelling-9ffef96a24f6#.x5xz09ogc">article about the character-based model for spell checking</a> for details. We used the same underlying approach and tweaked the model for higher accuracy.</p><ul><li><em>Accuracy results:</em></li></ul><p>We trained the model on Windows 10 machine with an Intel(R) Core(TM) i7–6700 CPU (2.60GHz) with Java(TM) 1.8.0_121. The best accuracy we were able to achieve was <strong><em>PreDict LSTM 90.71% accuracy </em></strong>after 8 days of training time. Overall we were quite impressed with the achieved accuracy even though we were not able to achieve our goal of &gt;95%.</p><ul><li><em>Performance results:</em></li></ul><p>Since query performance is crucial to our solution we measured the throughput queries per second to get some idea about the correction performance. On the same instance type, we where able to achieve up to <strong><em>PreDict LSTM thrpt 0,119 +-0,057 ops/ms</em></strong> (corrections per millisecond). This result was very disappointing. We know that we could maybe double or triple the performance by some fine-tuning and throwing more performant and expensive hardware on it but this is not our preferred solution. Therefore we searched for an algorithmic solution.</p><h3>2.The algorithmic way</h3><p>As already mentioned in <a href="https://medium.com/@searchhub.io/a-closer-look-into-the-spell-correction-problem-part-1-a6795bbf7112">Part 1</a> we are talking about approximate string search algorithms that allow to lookup a string in a list of strings and return those strings which are close according to a specific string metric. The best of breed algorithms all strive for the same goal to achieve short lookup times. This is mainly done by reducing the number of lookups and comparisons (between words and/or artificially generated candidates), possibly reducing further the number of full edit distance calculations and finally reducing the computational complexity of the edit distance calculation itself, while not compromising its accuracy.</p><p>We compared several algorithms that are said to be state-of-the-art in this area (<strong>BK-trees</strong>, <strong>FSTs</strong> and <strong>SymSpell</strong>)</p><p>When we found <a href="https://github.com/wolfgarbe/symspell">SymSpell</a>, we were really impressed by its internal simplicity, performance and accuracy. We thought it would be a good starting point to work on a solution with an even higher accuracy without losing much of its original performance.</p><blockquote><a href="https://github.com/searchhub/preDict">preDict </a>— is our open-source state of the art spell correction library solving the spell correction problem at scale based on SymSpell.</blockquote><p>Here we’re going to tell what we changed in detail and where we see opportunities to expand its capabilities.</p><p><strong><em>a. Refactoring</em></strong></p><p>Originally we forked from an existing Java fork, that looked like a horrible 1 to 1 translation from the original SymSpell. No blaming here — it worked pretty well as a POC. But simply by structuring it (it still does not fulfill many best-practice metrics), putting it into an object-oriented structure (which improved readability) and by applying some known Java tweaks, we were able to achieve a better performance.</p><ul><li><strong><em>PreDict CE thrpt 82,116 ± 1,149 ops/ms</em></strong></li><li><strong><em>Original SymSpell (Port) thrpt 68,105 ± 0,977 ops/ms</em></strong></li></ul><p><strong><em>b. Weighted Damerau-Levenshtein</em></strong></p><p>We changed the Damerau-Levenshtein implementation to support weights for single edit operations (insert, delete, replace, transpose). With these parameters we got the ability, to tune the library onto special test data with automated tests. This however should be done carefully, since test data rarely represent real use cases. It’s also useless if it should handle unknown or broad data. This is why we finally stayed with values near 1 (+/- 0.2).</p><p><strong><em>c. Keyboard Distance</em></strong></p><p>Today fuzzy errors happen most likely on mobile devices, where keyboards get smaller and people tend to write on the go. That’s where it happens quite often, that the finger hits the letter nearby instead the intended one. This is why we calculated a keyboard distance matrix and put it into our code. It is used in addition to the weighted Damerau-Levenshtein implementation when we ask for a replacement weight: if the replacement is a nearby letter, it gets an even lower penalty.</p><p><strong><em>d. Refined proximity for top-k results</em></strong></p><p>If the “accuracyLevel” (called “verbosity” at SymSpell) is set to “maximum” (2) and the “topK” value is set to a number greater than 1 (6 was pretty good), the last modification kicks in order to increase accuracy: We apply several proximity algorithms on these top-k results and combine them. The weighted proximity values are then used to reorder the final result.</p><p>We basically use SymSpell to pre-calculate several candidates and then prefer matches based on their phonetic proximity (using Eudex), prefix proximity (Jaro-Winkler) and a fragment proximity based on (Dice Coefficient).</p><ul><li><em>Accuracy results:</em></li></ul><p>We benchmarked the implementation on a Windows 10 machine with a Intel(R) Core(TM) i7–6700 CPU (2.60GHz) with Java(TM) 1.8.0_121. The best accuracy we were able to achieve was <strong><em>PreDict CE 90.04% accuracy </em></strong>after 2 seconds of indexing. It’s pretty amazing to see how good you can get even without a lot of data. <strong><em>In our Enterprise-Edition (PreDict EE) we use the same PreDict library and extend it with our “secret sauce”, that solves decomposition and add some more tricks which finally increases accuracy by other 7% to 96.86%.</em></strong></p><ul><li><em>Performance results:</em></li></ul><p>On the same instance type, we were able to achieve up to <strong><em>PreDict CE thrpt 86,019 +-1,127 ops/ms</em></strong> (corrections per millisecond). <strong><em>This is almost three orders of magnitude faster than the LSTM approach</em></strong>.</p><h3><strong>3.Summary</strong></h3><p>Based on our measurements it should be obvious that the algorithmic solution is the preferred solution for our spell correction problem. Beside the pure performance and accuracy benefits the algorithmic solution is way easier to deploy, operate and more cost efficient. But we wanted to do more by trying to combine the best of the two worlds.</p><p><strong><em>Bridging the gap between algorithmic and data-driven.</em></strong></p><p>Under the hood we built a hybrid solution to the spell correction problem that uses the algorithmic solution at runtime but uses deep learning to compute the individual weights / parameters for our highly parameterized algorithmic model. By doing so we were able to achieve even higher accuracy levels without impacting the throughput performance at all.</p><h3>We are hiring</h3><p>If you’re excited about advancing our search|hub API and strive to enable companies to create meaningful search experiences, join us! We are actively hiring for Data Scientists to work on next-generation API &amp; SEARCH technology.</p><p><a href="http://www.searchhub.io/">www.searchhub.io</a> proudly built by <a href="http://www.commerce-experts.com/">www.commerce-experts.com</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8993ecab7226" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A closer look into the spell correction problem — Part 1]]></title>
            <link>https://medium.com/@searchhub.io/a-closer-look-into-the-spell-correction-problem-part-1-a6795bbf7112?source=rss-de55c8251d7d------2</link>
            <guid isPermaLink="false">https://medium.com/p/a6795bbf7112</guid>
            <category><![CDATA[information-retrieval]]></category>
            <category><![CDATA[search-engines]]></category>
            <category><![CDATA[search]]></category>
            <dc:creator><![CDATA[searchhub.io Dev Blog]]></dc:creator>
            <pubDate>Fri, 11 Aug 2017 13:30:47 GMT</pubDate>
            <atom:updated>2017-08-21T09:46:19.097Z</atom:updated>
            <content:encoded><![CDATA[<p>At <a href="https://www.searchhub.io/">searchhub.io</a> our mission is to help search engines understand humans. You can see Search|hub as a search platform independent search query intelligence API that supercharges your existing search engine.</p><p>One important aspect in order to help search engines understand humans is to handle error prone human input aka spell correction. Let’s imagine an eCommerce context where users sometimes type the query <em>jewlery</em> when they’re looking for <em>jewelry</em>; sometimes they just accidentally hit an extra key and type <em>dresss</em> instead of <em>dress</em>. To make their online shopping experience successful, we need to identify and fix these mistakes and display search results for the canonical spelling of the actual intended query.</p><h3>How big is the spell correction problem?</h3><p>Let’s take the example of search and have a closer look at the distribution of different spelling errors we see on our API to get an idea about the size of the problem.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1004/1*5o3H5FWApyiMUkelzpfyUA.png" /><figcaption>frequency of spelling errors by error type</figcaption></figure><p>As you can see from the graph spelling errors are quite frequent and some of them have a significant influence on the value per search.</p><h3>Does context matter?</h3><p>But maybe an aggregated view on spelling errors is not enough. The likelihood of a spelling error and its type could also be related to context. One obvious example for a context variable would be the device type.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1004/1*nxCPZWTfV4CZtWmKm_vIew.png" /><figcaption>frequency of different spelling errors by device</figcaption></figure><p>Again we can directly see some differences between the distributions for desktop and mobile especially when it come to replaces. This makes perfect sense since the keyboard on a mobile is so small that the probability of hitting the wrong key is significantly higher than on larger keyboards.</p><h3>What about stemming, lemmatization and term decomposition?</h3><p>But wait — maybe that’s still not detailed enough. As you can see from the graphs we also have to take care of term decomposition, stemming and lemmatization. Most of these areas can be handled by dictionary approaches but all of them fail miserably if they have to deal with a combination of a spelling error and one of these areas. ( for example “<em>womensaparel</em>” to “<em>women’s apparel</em>”)</p><blockquote>It becomes pretty clear that any system (search engine, speech recognition, chatbot, etc.) that is using direct unfiltered human input but fails to invest in best-in-class spell correction technology leaves a significant junk of money on the table.</blockquote><h3>Defining the Problem</h3><p>Now you could argue that the problem of spell correction is already solved as it is well understood and a lot of work has already been done in the last couple of centuries but we think it is not or it wasn’t :-)</p><p>The problem is directly related to approximate string matching / string distance which allows to lookup a string in a list of strings and return the ones that are “<em>most similar</em>”. There are are many different string distance metrics like <a href="https://en.wikipedia.org/wiki/Levenshtein_distance">Levenshtein</a>, <a href="https://en.wikipedia.org/wiki/Damerau%E2%80%93Levenshtein_distance">Damerau-Levenshtein</a>, <a href="https://en.wikipedia.org/wiki/Hamming_distance">Hamming distance</a>, <a href="https://en.wikipedia.org/wiki/Jaro%E2%80%93Winkler_distance">Jaro-Winkler</a>, <a href="http://www.catalysoft.com/articles/StrikeAMatch.html">Strike a match</a> and others out there but in our humble opinion they are not good enough mainly because they are oversimplifying the problem space. Therefore we asked ourselves the following fundamental question:</p><blockquote>How can we measure similarity and is it a single dimensional problem or do we have to take multiple dimensions into account?</blockquote><p>We think that similarity has to be multi-dimensional because the causes of spelling errors are multi-dimensional as well.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/907/1*PoTt2Cgeueqc68yO6Y8Czw.png" /><figcaption>different error-types in spelling errors</figcaption></figure><p>Not only you have to incorporate this multi-dimensionality, you should better use weighted string distance algorithms as well. It is not enough to use simple unweighted edit-distance algorithms to spell correct human input. There may exist thousands of strings that share the same edit distance (<em>button</em> vs. <em>butter</em>) or even a smaller edit distance (<em>mispeld</em> vs <em>mislead</em>) to the correct string but are completely different in their meaning. Without edit weights based on real user error probabilities you’ll end up with wrong spell corrections and frustrated users.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/963/1*xg-sE7KtOFZ5ejMQbZpbtQ.png" /><figcaption>when simple edit distance goes wrong</figcaption></figure><h3>Manual mapping can’t fix misspellings at scale</h3><p>We often see clients manually mapping misspelled words to the correctly spelled word by adding synonyms. But this means hours of manual effort and intervention. While this might be a great solution to solving small scale problems, it is highly ineffective if you have a large catalog with thousands of products with an very dynamic assortment, in which people search using complex compositions of words and symbols.</p><p><em>In our next blog post we’ll share some insights on how we tackled the spell correction problem at scale in search|hub with unmatched speed and accuracy.</em></p><h3>We are hiring</h3><p>If you’re excited about advancing our search|hub API and strive to enable companies to create meaningful search experiences, join us! We are actively hiring for senior Java DEVs and Data Scientists to work on next-generation API &amp; SEARCH technology.</p><p><a href="http://www.searchhub.io/">www.searchhub.io</a> proudly built by <a href="http://www.commerce-experts.com/">www.commerce-experts.com</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a6795bbf7112" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Introducing search|hub.io — super charging your search engine]]></title>
            <link>https://medium.com/@searchhub.io/introducing-search-hub-io-super-charging-your-search-engine-5df6395b4664?source=rss-de55c8251d7d------2</link>
            <guid isPermaLink="false">https://medium.com/p/5df6395b4664</guid>
            <category><![CDATA[search]]></category>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[naturallanguageprocessing]]></category>
            <dc:creator><![CDATA[searchhub.io Dev Blog]]></dc:creator>
            <pubDate>Tue, 20 Jun 2017 07:09:00 GMT</pubDate>
            <atom:updated>2017-08-04T19:52:51.042Z</atom:updated>
            <content:encoded><![CDATA[<p>last week we officially launched our new baby called <a href="https://www.searchhub.io/">search|hub.io</a> coming directly out of our 6 months stealth mode. After having successfully optimized roughly 50.000.000 site search queries, spending 15 months of development, going through 5 live integrations and 3 incredibly won A/B tests on our customers live environments we decided that it is time to tell the world what we strive to do.</p><blockquote>“Our motivation “FAST &amp; RELEVANT SEARCH” helps to craft meaningful search experiences for the people around us, this mission inspires us to jump out of bed each day and guides every aspect of what we do.”</blockquote><p><strong>some background…</strong></p><p>Everybody in our team of world-class engineers and data scientists has a very strong background in NLP and Information Retrieval. All together we spent more than 50 years in the domain of e-Commerce Search working for the leading pack of software vendors and retailers in this space in order to make shopping easier and more relevant for the 1.77 billion people shopping online. 18 months ago having worked through several big site-search re-platforming projects with some of our customers we finally realized that in order to <strong><em>make search better and ready for the future </em></strong>we have to do it on a completely different level than before.</p><p><strong>the role of search…</strong></p><p>Search Engines essentially act as filters for the wealth of information available on the Internet. They allow users to quickly and easily find information that is of genuine interest or value to them, without the need to wade through numerous irrelevant web pages. However, they are built based on concepts that stores words and not their meaning. Screen Real estate is going to zero and the way people can interact with digital systems has evolved from Keyboard only to Keyboard, voice, images, and gestures. All of these input types are error prone and don’t include context. <strong><em>Well done search, however, should cater for these errors and automatically add available context to come up with relevant answers.</em></strong></p><p><strong>our entirely new approach with search|hub.io…</strong></p><p><em>“search|hub is a search platform independent, AI-powered search query intelligence API specifically designed to enhance our customers existing search, not replace it.”</em></p><p>We really like this description since it pretty much perfectly represents how we think about the future. But let’s start at the beginning of the sentence:</p><ol><li><strong>search platform independent..</strong></li></ol><p>…during our past we’ve seen a lot of different search engines out there and all of them have their pros and cons. Often the reason why a specific search engine was chosen has nothing to do with the ability to successfully fulfill the basic use case of search. In more complex scenarios there could even be a set of different search engines in production each fulfilling a different use case.</p><blockquote><em>That’s why we designed search|hub as a Proxy that automatically enhances your existing search engine(s) by autocorrecting, aligning human input and building context around every search query in a way your search engine understands.</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*i2u568zQmF2gfth00Ym00w.png" /><figcaption>search|hub’s search platform independent architecture</figcaption></figure><p><strong>2. AI-powered search query API…</strong></p><p>…various forms of AI have already begun to impact the quality of search results and the accuracy with which search engines interpret search queries. More important AI will further impact search in several ways:</p><p>It will impact the way search engines understand search queries. Search engines will slowly transition from responding to queries by matching the words contained in the query to the words that appear on website pages, to responding to queries based on a deeper understanding of what they mean. It will impact the way data is gathered, interpreted, and structured, making the process of searching more efficient and effective, and more conform to what people actually want to know. In this sense, search engines will begin to accumulate ‘knowledge’ which will allow them to provide more intelligent responses to people’s search queries. It will impact the way Search works and the way we experience it. Search can be so much better than it currently is. It can include conversational elements in it, in which you can actually have a dialogue with the intelligent and knowledgeable engine you are conversing with. It can include highly personalized and contextually aware suggestions and recommendations. It can include contextual data intelligently amassed from the web at large. It can provide expert advice (based on its knowledge) in the form of a virtual expert with which you can converse, naturally, until you find what you were looking for and are happy with it.</p><blockquote>AI helps search|hub to automatically optimize search queries by correcting spelling errors, typos and other forms of errors with unmatched accuracy &amp; consistency and speed. In a second stage, it uses a unique combination of NLP &amp; Deep Learning to optimize every search request by correctly identifying search intent and maximizing relevancy. As it gathers behavioral data it further refines the results for even greater relevance.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-eiIrsHVqER_VP9yhdiuHA.png" /><figcaption>search|hub’s smart query rewrites</figcaption></figure><p><strong>3. enhance, not replace</strong></p><p>…search Engines became an integral part of modern system backbones and are in most cases very hard to replace. But “search” as we know it is already changing because modern technology has given today’s customer’s very high expectations; they know what they want and they want it now. And while today’s customers use the search bar, by tomorrow they could be using voice and image search or even combinations of that. However, to perfectly serve their needs in order to grow your future business you need to understand them.</p><blockquote><em>The technology behind search|hub was specifically designed to enhance our customers existing search, not replace it. With just two API calls, search|hub integrates as a proxy between your frontend application and your existing search engine(s) injecting its deep knowledge.</em></blockquote><p>If you’re excited about advancing search|hub technology and enabling companies to create meaningful search experiences for the people around us, join us! We are actively hiring for senior Java DEVs and Data Scientists to work on next-generation API technology.</p><p><a href="https://www.searchhub.io/">www.searchhub.io</a> proudly built by <a href="http://www.commerce-experts.com/">www.commerce-experts.com</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5df6395b4664" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>