Leveraging Social Media Content for detecting Insurance Fraud Using IBM Bluemix

Transcription

1 Leveraging Social Media Content for detecting Insurance Fraud Using IBM Bluemix Ramanathan K Enterprise Information Architect Ramanathan.k@tcs.com Ruchi Agarwal ATU-Bluemix Lead Noida ruchi.agarwal1@tcs.com Nita Khare ATU Global Lead, IBM Analytics Mumbai , Nita.khare@tcs.com ABSTRACT Long gone are the days, in which organizations across the world analyzed only the data that they had in their transactional systems or relying on human experts only for detecting Insurance Fraud. With the rise of social media and people willing to share everything about their life in open in those social networks knowingly or unknowingly, Insurance organizations across the world suddenly have got access to a big treasure trove of data to better understand their customers and their associations. This paper talks about leveraging this new glut of data found in the social networks for the advantage of an Insurance organization to detect and thus prevent fraud before it happens. IBM s cloud based Bluemix platform with its rich repository of Watson services has been used to develop this solution. 1. INTRODUCTION One of the key challenges for Insurance organizations across the globe is early detection and management of fraudulent claims. One of the studies has shown that approximately 10% of the claims reported to an Insurance company could be fraudulent. This affects the company s bottom line severely and in turn affects the policy holders as well in terms of higher premiums for all. In this paper, we will focus only on the Insurance Fraud Management Use case and see how easily the frauds can be detected and prevented using TCS IBM Insurance Bluemix Solution. vulnerabilities of the Insurance companies. Moreover these tools and techniques worked well detecting individual claims fraud. Then came the social media era, in which the customers started sharing anything and sometimes everything about their personal lives openly with their close associations easily visible to all. Now with this additional data to supplement the internal transactions data, finding organized fraud rings with various stakeholders involved in a claims settlement is becoming possible. Now with the veracity of the social media content being in question and the various privacy laws preventing the usage, this data can still be used to raise the flags to the Special Investigation Units(SIU), fraud analysts and claims managers who in turn can then take the investigation to the next level. 2. Possible Insurance Fraudulent Scenarios Customers across the world are extensively using the social media to update their status, moods, likes and dislikes from wherever they are. They do this with their network which consists of their associates, colleagues and acquaintances. The TCS IBM Insurance Bluemix solution leverages the sentiment analysis of the content, the location from which the post was generated, presence of keywords leading to fraud and the follower/network list of the customer to detect fraudulent patterns and raise flags to the SIU. 1.1 Insurance Fraud Management Historically Insurance companies have been relying on human experts to understand fraudulent patterns and investigate them with most of them being done retrospectively much later after the claims were paid. With the advent of computer systems, Insurance companies started using the data within the organizations to detect anomalies by implanting business rules in their applications to alert them for fraud patterns. However these rules were limited by the available data and hence were not completely effective to detect the fraudsters who keep changing their tricks to exploit the Copyright 2015 Page 1 of 6

2 3. IBM BLUEMIX OVERVIEW Figure 1 Claims Fraud Analysis Sentiment analysis of content Events leading to Insurance claims are almost always not a very pleasant experience and hence carry negative perspectives at least a few days before and after the day of the claim. Hence excessive happiness expressed in the social media during these days often is questionable but not impossible. The objective here is to do a sentiment analysis of the posts on the eve of the event and mark excessively positive sentiments for further analysis. Location of origination of the post All the claims have a location associated with the events leading up the claim. The objective here it to ensure that the location of the social media post does not contradict with the location of the event mentioned by the claimant while submitting the claim. For example, the location of the post cannot be a restaurant or a totally different city/place at the time mentioned as time of accident in the claim submission forms. Presence of fraudulent key words The objective here is to identify key words (related to fun/travel or sports) which indicate potential fraud for example a person enjoying in a local bar or a dance show or a sports activity at the time of the event or just a few hours after the event. Analysis of follower/network There are multiple parties involved in a particular claim including doctors, body shop owners, adjustors, witnesses at the scene etc. and it is highly unlikely that these people know each other all the time. The objective of this analysis is to find if these seemingly unrelated people are conspiring to form an organized crime ring. Many types of web applications are running on the Internet today. There are also many ways to manage and maintain the infrastructure that powers those applications. IBM Bluemix delivers quick and easy cloud capabilities to deploy and maintain a web application, with minimal hassle and overhead. Figure 3 IBM Bluemix Overview Bluemix is an open- standard, cloud based platform for building, managing, and running applications of all types (web, mobile, big data, new smart devices).. In addition to providing additional frameworks and services, Bluemix provides a dashboard to create, view, and manage applications and services as well as monitor application s resource usage. Bluemix is an end-to-end offering that provides a complete set of integrated services to simplify development, test, build, and deploy applications. Furthermore, Bluemix provides a rich set of services to extend an application through incorporation of sophisticated analytics, social, and mobile capabilities as needed. Finally, the Bluemix platform frees developers and organizations from worrying about infrastructurerelated plumbing details and focus on what matters to their organizations: business scenarios that drive better values for their customers. Copyright 2015 Page 2 of 6

3 3.1 IBM Bluemix Services elasticity and agility of the cloud with a highly secure infrastructure. IBM Bluemix offers extensive catalog of services. It includes IBM as well as third party services. Below diagram shows a snapshot of different service categories available as of date. Figure 5 DashDB Service Figure 4 IBM Bluemix Services Following are some of the key services which were used in developing the solution described below in this paper Insights for Twitter Insights for Twitter service helps to incorporate Twitter content from the Twitter deckhouse into the Bluemix applications. The content store is refreshed and indexed in real-time, making searches dynamic and fast. Sentiments and other insights can also be applied on twitter content using this service Alchemy API: Alchemy API uses natural language processing technology and machine learning algorithms to extract semantic meta-data from content, such as information on people, places, companies, topics, facts, relationships, authors, and languages. It provides easy-touse mechanisms to identify positive/negative sentiments. Figure 6 Insights for Twitter Below are the other Bluemix services used in the solution: 1) Pitney Bowes-It help to track the location of the user in real time. 2) Dataworks to refine the twitter data and make it suitable for analysis 3) Twitter API to connect to twitter feed 4. TCS IBM BLUEMIX FRAUD SOLUTIONS Figure 4 Alchemy API Dash DB DashDB is a fully managed cloud data warehousing service.dashdb helps extend our existing data warehouse appliance into the cloud, or build entirely new, self-service data warehousing infrastructure. It is powered by high performance inmemory and in-database technology that delivers answers very fast. DashDB provides the simplicity of an appliance with the A key challenge for insurers is to strike a balance between fraud detection and claims handling demands. The solution described below offers a comprehensive and streamlined approach to fraud detection which helps to increase the detection rate using social data like Twitter, Facebook etc. The solution leverages advanced analytics to analyze the claims and has the potential to help insurers worldwide to efficiently and accurately detect more fraudulent activities. Copyright 2015 Page 3 of 6

4 4.1 Case Study ABC Insurance Company Scenario Following is an example fictitious case from ABC Insurance Company to explain the solution. Customer John submitted an accident claim to ABC Insurance Company stating that he met with an accident on 24-Sept-2015 at 10:30 AM in Hammond City of Indiana State in US. He mentions that both his legs have got bone fractures in the accident. His doctor s name is Ms. Elizabeth O Shea..John s tweets, a few days before and after 24 th September is collected for analysis. Below is a snapshot of John s claim: Figure 7 John s Twitter Sentiments Below is a sample snapshot of John s tweets to explain the above score: Figure 8 John s sample data TCS IBM Insurance Solution has different types of analytics to detect and flag fraud via an integrated approach of consolidating information from various social networking sites. Each type produces a meaningful result, however, when combined, increases the quality of results. The current solution provides insurers with 4 types of analytics to uncover hidden complex fraud patterns primarily using Twitter feeds. However the solution is extensible for other social networking sites as well. Following is a detailed explanation of each of those types of analytics Sentiment analysis of content Events leading to Insurance claims are almost always not a very pleasant experience and hence carry negative perspectives at least a few days before and after the day of the claim. TCS IBM Insurance solution will leverage this general psychology to analyze sentiments of the social media content before and after the claim is raised. For the given scenario, solution provides the following results: 1) John s tweets result in an overall positive sentimental score. Figure 8 John s tweets sample Below is the sample flow in the TCS IBM Bluemix solution to come up with this sentimental score. Figure 10 Sample Flow Location of origination of the post Each claim has an associated location and the idea here is to verify if that location does not conflict with the location information from a Tweet. For example The accident location and hospital location cannot be too far away from the location of the tweet just a day before of after or on the day of the claim. For the given scenario, solution provides the below results: John s location based on social site is Benda City, Indiana which is different from his accident location on the day of his claim. Copyright 2015 Page 4 of 6

5 Below is the sample flow in the TCS IBM Bluemix solution to detect the key words indicating a possible fraud Figure 12 John s Accident Location Below is the sample flow in the TCS IBM Bluemix solution to detect the location of the tweet. Figure 13 Sample Flow Presence of fraudulent key words TCS IBM Insurance solution also helps to mine information from social networking portals looking for keywords belonging to different categories like Sports, Fun, Travel etc. It helps the insurer to verify if the claimant participated in any sports or fun activity or travelled elsewhere on the same day or just after the accident. It also assists in verifying the intensity of accident with the claim amount. For example: a person enjoying in a local bar or a dance show or a sports activity at the time of the event or just a few hours after the event is highly unlikely to have undergone a serious accident Figure 15 Sample flow Analysis of network of the claimant: TCS IBM Insurance solution helps to analyze the relationships and connections among the various participants of the.for example, if the claimant has the doctor, chemist or the body shop owner in his network list, there is a possibility that they might have conspired to cheat the Insurance Company. For the given scenario, solution provides the following results: It was found that John s doctor was indeed one among his twitter followers For the given scenario, solution provides the following results: 1) Few Keywords like: Sports, Travel, Celebrate, Party, Night out are all detected from John s tweets. 2) He travelled to Victoria Peak on the same day when his accident happened. Figure 16 John s doctor as his twitter follower Each of the above analytics type could set a flag or a comprehensive score can be calculated based on the number of flags which can then be passed on to the SIU along with the detailed information. The SIU can then prioritize the cases with more flags or score and move on to do a detailed analysis to verify the facts uncovered by the social media content analysis. 5. BENEFITS Expanding the fraud detection potential Bluemix solution gives a broader and additional view by processing social media content through advanced analytics. Organization can detect linked entities and crime rings to stem potential losses at an early stage. Figure 14 John s Sample tweets Lower claim processing expenses Copyright 2015 Page 5 of 6

6 Organizations can effectively reduce false positives/trigger flags for the undetected ones using the above Bluemix solution. Thus they can help to improve investigator efficiency and conduct more efficient and accurate investigations Gain a greater competitive advantage. Effective fraud detection methods can eventually discourage fraudsters from targeting the organization and thus help in gradual reduction of fraudulent claims Potential to scale the solution The solution can be expanded to other/newer social networking sites and can also process sudden/increase in volumes of data as it is cloud based. Thus scalability and flexibility are major advantages of this solution. 6. Future Roadmap Mentioned below are few other use cases in the roadmap for the TCS IBM Insurance solution to expand in to areas other than fraud monitoring.. Brand Reputation Monitoring - Understand how a particular brand is performing by measuring reach, activities, sentiments, trends Competition Analysis - Analyse how competitor's brand is performing under specific themes Customer Digital Engagement Measurement and Monitoring - Measure how digitally engaged the customers are, by measuring involvement, trends, influence score etc. Customer Frequent Queries analysis - Analyze frequently asked queries, issues raised, sentiments around resolution to identify process improvement, training needs Influencer Identification - Identify who are strong proponents of the brand and to leverage them for improving the brand reach and reputation Understanding trends/events in P&C business - real time monitoring of fatalities/global events/catastrophes/flu breakouts which could be of impact to Insurance business Real time risk monitoring and adjustments to reserves 7. Summary The financial impact and incidence of fraud are on the rise in insurance companies, due in part to the fact that today s fraudsters are harder to detect. They are acting much like legitimate customers. They are also highly organized and rely on increasingly sophisticated criminal tactics and masterful exploitation of organizational vulnerabilities.handling fraud manually has always been costly for the insurance companies, even if one or two high value fraud went undetected. The solution described above helps to expand the potential of fraud detection at the same time reducing the overall cost of fraud detection by leveraging social media content. Social network analytics has proven particularly beneficial in the fight against insurance fraud. Using the Bluemix solution, insurer can take a hybrid approach that combines the use of business rules with internal enterprise data and usage of advance analytics, keywords, extraction, sentimental analysis, location of fraudsters as well as network linkage analysis with external data to effectively uncover those previously hidden issues. The solution has the potential to scale by including additional social networking sites, additional scenarios of fraud as well as additional volume of data for analysis as it is on a cloud based platform. Also when the future roadmap indicated above is fully developed, this could become a game changing solution for an Insurance Organization. 8. ABBREVIATIONS Abbreviation SIU 9. REFERENCES Table 1. List of Abbreviations Description Special Investigation Unit Copyright 2015 Page 6 of 6