Showing posts with label Big data. Show all posts
Showing posts with label Big data. Show all posts

Sunday, December 24, 2017

AI in Recruitment: Word2Vec Opens up Interesting Possibilities

CVs and jobs are text heavy and like all the challenges which exist with natural language – multiple ways of describing the same concept, ambiguity, synonyms etc.  Meaningful interpretation of text requires extracting this knowledge in a machine understandable form. Among others similar problems exist in speech recognition, machine translation and conversational systems like Siri.

AI systems that process images work on high dimensional vector representation for each pixel embedded in a two-dimensional image. Most of the information needed to recognise images is present in the two-dimensional vectors.  However, most text processing system use a “bag of words” representation of text, that is each word is represented by a ID. For example, Infosys and TCS may be represented as say, ID75698 and ID 98603. And don’t use the contextual relationship between the two words, which otherwise recruiters or jobseekers can understand and process.

Latent Semantic Analysis is a technique which condenses the statistical count of co-occurring words into topics or concepts. It has been shown that Latent Semantic Analysis would recognise a shallow kind of topical similarity and not work well where subtle semantic relationship between words is present.

In contrast, predictive methods like Word2Vec learn the function that captures the salient statistical characteristics of the distribution of sequence of words. The function can associate each word with a continuous-valued vector representation that corresponds to a point in a feature space.

Word2Vec takes raw text as an input and the training of the Word2Vec model (skip-gram) is to arrive at vector representations of words that best predict a window of surrounding words. One can imagine that each dimension of that space corresponds to a semantic or grammatical characteristic of words.

The hope is that similar words get to be closer to each other in that space- that is we may expect Infosys Technologies and TCS as companies to be much closer to each in this feature space. That opens up new possibilities for AI in recruitment.

Thursday, January 3, 2013

Story of Naukri Job Alerts


Naukri.com is the market leader among with career sites in India, with market share currently at 63%. Naukri.com has 30 million+ registered profiles and a large part of these registered members receive a job alert every alternate day or on a weekly basis. Job alerts only contain freshly posted jobs on Naukri.com in last two/three days. It is probably the main reason why Naukri Job Alerts have one of the highest open and click through rates. Yet, jobseekers complained of relevance of jobs sent. That was identified as one of the important problems to address in early 2010.

The process of improving the job alerts was an incremental one. We built the logic step-by-step and with every incremental step, our understanding of the relevance problem improved.

1. Discovery of “Role” – I tend to believe one major variable than we discovered was “Role”. A deep dive in the behavioral data showed several interesting patterns. Jobseekers were not sticking to their Functional Areas (departments) and were applying across Functional Areas.

a. Pattern of apply clearly indicated that Role was more important than functional area.

b. We had roles which were very similar present in multiple Functional Areas, for example, sales role existed in Industry specific functional areas. GM Accounts existed in Accounts Functional Areas as well as the Top Management Functional Area. Also, several functional areas were close to each other, for example, Accounts and Banking.

2. Limitation of Keyword search – Key skills entered by jobseekers represents what they consider as important. Logically, a search on jobs should use the key skills entered by the jobseeker. However, some of the jobseekers had not entered their key skills. A large gap existed in the key skills entered and their skills as apparent from the CV. We needed a robust mechanism which did not fail because of the data inadequacy.

3. Handling of Categorical Variables – When we compare two jobs and their relevance to the jobseeker, attributes like “role” were important. The key challenge was to translate this into a distance function that can be used in predicting relevance for the jobseeker. Similarly, attributes like Industry, Location required identification of a good distance function.

4. Jobseeker Resume – A match between a jobseeker’s expertise and the requirements from the recruiter is essentially a match between the CV/resume and the Job Description. Of course, there are challenges – if a CV is old or a job description is incomplete, this may not work very well. Yet, we needed a mechanism for matching the candidate CV and the job description.

5. Apply Behavior - It is very much possible that apply behavior of a jobseeker will deviate from the CV/resume.  Apply behavior can provide insight into asiprations of the jobseekers as well as help identify classification errors. Incorporating apply behavior in identifying matching jobs for jobseekers is another significant challenge.

Naukri Analytics team identified the above challenges and incrementally solved them in association with the product team and the technology team. And of course, we noticed a major improvement in relevance feedback from jobseekers.

We are not done on solving this technical challenge. Analytics team is looking to hire smart Data Scientists to join its rank and work on solving these – if you are interested, please click to here to apply.

Tuesday, August 21, 2012

Analytics -Making the Transition From “I Feel” to “I Know”

A young, bright, energetic MBA graduate from a top college joined the “Most Innovative Company Pvt Ltd”. He was full of ideas and keen to change the world with his overflowing cup of knowledge. Within one month of joining his company, he presented an interesting idea to his manager. His manager asked him to learn more details about the company, the business context and prepare a proposal. However, by next week, manager was running after quarterly targets and he asked the new joinee to focus on the targets, rather than the new idea he was pondering over.

Three months in the company, manager asked the trainee to focus on another idea, manager’s pet idea, which he has been pushing for a few years and was not able to move forward for lack of time. The trainee was assigned the task of preparing a presentation, putting the stuff together and do the background work. He got to spend a lot of time of his manager and three months down the line, the proposal was ready.

Since Business Head was visiting for annual review, his manager presented the idea to the Business Head. Business Head liked the presentation but did not feel that idea was significant. He asked the manager (and in turn, the trainee) to work on another project which was Business Head’s pet project. Another six months down the line and much effort spent, even that idea was rejected because CEO did not feel that the idea will succeed.

Does the story sound familiar? A lot of ideas get rejected because the idea champions fail to convince the decision maker. If a decision maker FEELS that idea will not work, that is the final outcome. If you look at the whole story, it was all about different people making decisions on what they FELT will work or not. In the end, idea champions failed to present data and customer insight to support their idea. Projects were conceived because managers FELT that they will work, they did not KNOW that they will work.

That’s where analytics is important, in removing the guesswork from decision making. It is important to use customer insights from data and if data is missing, collect it before making the pitch. Make the decision maker believe that you KNOW that the project will work.

Friday, August 17, 2012

Big Data and Analytics

Data Analytics is as old as the data itself. Extracting information from data is as ancient as one can think of. However, there is a major hype around Big Data, Web Analytics in the recent past. What has changed–
(a)   Most businesses across the globe have implemented some form of ERP and transaction automation solutions. Customer Relationship Management (CRM) solutions have been in existence for more than a decade and a large number of organizations have crossed the bridge in terms of integrating different systems across the organization. It is now practically feasible to understand a customer in terms of her journey across the organization from a lead, to sales pitch, to customer, service/product delivery and fulfillment.
(b)   Transactional data is large in volume, often captures the demographic information and behavioral information in terms of likes (products purchased represent a Like behavior). Large volume of data naturally can be used to identify patterns and patterns can be used to characterize customers. For example, a customer purchasing lot of sugar free items is possibly health conscious or may be diabetic. Customer buying lot of sports goods is obviously interested in sports.
(c)    Analyzing Retail transaction data and interpreting basis the purchase history was subject of intense analysis in Research in the Nineties. Since then Ecommerce has provided additional input from data analysis perspective. Ecommerce companies can track the browsing behavior of customers on their site. Which products they looked at, compared and what did they finally purchase.  Now this is enough to characterize customers as price conscious, style conscious etc. While transactional data can be used to derive similar inferences, click data is more definitive as it captures the customer’s exploration of products prior to making a decision. In the end, objective of all analysis to drive business decisions which can help the customer as well as the business.
(d)   It’s been more than a decade of big data and analytics in the Western world and Indian companies are now catching up. ERP which was big in early part of the last decade is now commonplace. Consultants are now keen on selling Analytical Tools and making a pitch of analytics. In the end, transactional automation solutions were also sold with the promise of data analytics. Customers are also demanding results from the same vendors.
The hype around big data and analytics is probably here to stay, unlikely to go away in a hurry. If a business analyzes several projects it undertakes, it does not take too long to realize than RoI from analytics projects is probably the highest.  As one of my friends once said “If you did not find the patterns in data that you were looking for, you did not look hard enough”. Data always throws patterns, patterns drive understanding of customer behavior, and hence, better decision making.