Hachi helps you in leveraging your combined network (Facebook, LinkedIn, Gmail, CRM contacts, etc.), to reach out to people who matter to you – be it for professional, social or personal reasons. We do a holistic search across all your social and professional networks, and we optimize the results based on relevance. Like most search sites, you can search using the top search bar, or the advanced search feature where you can search category-wise.
Top search box
We noticed that very few users used the “advanced search” feature. Google searches have made us lazy, and we expect the website to figure out what are looking for, and search category-wise, and give us the results. Most good sites will give you relevant results, but you will also get a large number of unwanted results. If you search for “Jeff, New York”, you might mean “Jeff who lives in New York (currently, or in past)” , and not the “Jeff who has New York in one of his job descriptions”.
Our goal is to give you the most pertinent results, instead of large number of unwanted results, and to make semantically relevant content appear on top. To achieve this, we implemented a self-improving algorithm which mainly does two things:
- Disregarding keyword count while ranking the results.
- Search categorization
Disregarding keyword count while ranking the results
“Searching for people” is different from “searching for articles”, because an individual has properties like location, title, profession, etc. At Hachi, we believe that the “query keyword” count should not be a metric for ranking search results.
We noticed that on LinkedIn, many users hacked their search ranking by including certain keywords multiple times in their profile. Mentioning a keyword like “SEO expert” multiple times in their profile doesn’t make a person a better “SEO expert” than a person who mentioned it only once. Similarly, a person mentioning her company name multiple times in her profile doesn’t make her profile more relevant than the person who mentioned it only once. Both work in the same company, and both should be given equal search ranking in this regard.
We give boolean scores to profile properties. A profile can have a certain property (or not) – and based on that, a score of 0 or 1 is given. If we cannot establish this, the probability of having that property is computed. For instance, a person mentioning “iOS developer” multiple times in her profile is given the same score as a person who mentions it only once. Both get an equal search relevance when “iOS developer” is searched.
Here, we simplified the search for users by automatically categorizing the query instead of asking users to do so. We tried to guess what the user “meant”, to give her the most relevant results. We used the existing data corpus to classify the query into categories.
- “IIT Delhi” can be classified as both “IIT Delhi[Education]” and “IIT[Education], Delhi[Location]”. The algorithm which we implemented categorized “IIT Delhi” as “IIT[Education], Delhi[Location]” and “IIT Kharagpur” as “IIT Kharagpur [Education]” where Delhi and Kharagpur are both locations.
- Christina from Cambridge mostly means Christina who studied at Cambridge[Education] (even if Cambridge is a location)
- John from Minnesota mostly means John who is from Minnesota[Location]
Categorizing ambiguous queries is difficult. When user searches IIT Delhi, it can mean both “IIT Delhi[Education]” and “IIT[Education], Delhi[Location]”. We devised a scoring algorithm based on the existing corpus, and that algorithm is handling the categorization here. This algorithm categorized “IIT Delhi” as “IIT[Education], Delhi[Location]” whereas a similar query “IIT Kharagpur” was classified as “IIT Kharagpur [Education]”. Now here, Delhi and Kharagpur both are locations. So, why the difference in categorization here? As per the current logic, there are more number of people who studied at IITs and are living in Delhi, so Delhi as a location gets more score as compared to “IIT Delhi [Education]”. On the contrary, very few live in Kharagpur, so “IIT Kharagpur [Education]” got a higher score. We feel there’s scope for improvement here, and we would love to hear your views on this. Do share!
You can check out these and a few more improvements in our search algorithms, at http://www.gohachi.com/. Please do send in your comments, suggestions and ideas at firstname.lastname@example.org. We would absolutely love to hear from you.🙂