Tigerobo: Building the next-generation search engine with natural language processing

Tigerobo CEO and founder Chen Ye © Tigerobo

With the success of Tigerobo Search, its flagship AI-based finance industry search engine, the startup is also diversifying into government, energy and media sectors

Imagine you want to know how many Tesla Model 3 cars were produced in the third quarter of 2019, and so you type the keywords into Google or another popular search engine. The results are likely to be a list of URLs. After sifting through several links and trawling a number of source articles, you may find a news report that gives the answer as “nearly 79,900."

Shanghai-based startup Tigerobo offers an alternative. Its AI-based smart search engine, Tigerobo Search, instantly presents a table extracted from a 28-page PDF document released by Tesla, which shows the answer as “79,837.”

Now the company is building a next-generation search engine based on natural language processing (NLP) technology that understands plain language questions – asked to chatbots and in future, smart speakers – and responds with precise, structured answers, instead of only giving users the URLs.

“It’s time for search engines to evolve,” said Tigerobo CEO and founder Chen Ye in a recent online interview with CompassList, and joined by CFO Kelly Liang.

During his long, successful career in search advertising for tech giants in the US and China, including Yahoo, eBay, Microsoft and Meituan-Dianping, Chen noticed significant advances in NLP technology from 2017–2018 based on a widely accepted test called SQuAD. Hosted by Stanford University to monitor the progress of NLP development, the bots used in the SQuAD test would read Wikipedia articles and answer questions based on their understanding.

By 2017, NLP accuracy was around 70% and in 2018 it reached 86%. Chen was convinced that once SQuAD accuracy reached 95%, NLP bots could be the primary tool used by human beings to acquire information. Meanwhile, Chen also identified the growing need for efficient data searching in certain sectors such as finance. So in 2017, he and his team started developing Tigerobo Search to respond to plain, natural language questions. 

95% SQuAD accuracy

Tigerobo Search uses semantic models and knowledge graphs to teach its algorithm to understand internet content and provide accurate answers. For example, if a user wants to know which companies had pivoted to producing masks and ventilators during the Covid-19 pandemic, Tigerobo Search can provide an exact answer because it has trained its algorithm with knowledge graphs of the medical equipment industry and supply chain.  

The startup created industry-specific knowledge graphs by extracting and aggregating information, including data on entities, names, relationships and figures from unstructured documents like news reports. Fed with such semantic models and knowledge graphs, the algorithms can instantly understand the question and locate related companies.

Tigerobo Search, which the company says has achieved 95% SQuAD accuracy, is able to give users a direct answer along with URLs via its proprietary algorithm based on NLP technologies. Users can choose to manually sift through those URLs for more related information or another result just in case they think the current one is not accurate enough.

The company also developed technology to crawl data from sources in different formats, including PDF, XLS, image and audio, because data, especially from its finance industry customers, is also buried in research reports published in such formats. It has added a translation feature based on machine learning to provide information in the major languages. 

The company has 80 programmers in its tech team, which make up four-fifths of its staff.

Bigger budget for tech solutions

When Tigerobo Search was launched in 2018, the startup targeted the financial industry, helping users search information on companies in the primary and secondary markets, bonds, funds, commodities as well as the economy as a whole. There are many reasons for this. 

Financial information is mostly structured and statements come in a fairly standard form and are freely available online, which made it easy for Tigerobo to acquire and use the data to train its algorithm, said Liang, the CFO, who like the CEO last worked at Meituan-Dianping, where she was Associate Director. “Whether a company is listed in the US or China, it is obliged to disclose certain financial information regularly."

She added: “Even private companies are required by authorities to comply with financial information disclosure requirements. But it’s very different when it comes to the medical or education industry where information is not freely available online due to privacy reasons.” 

Another reason is the finance industry’s willingness to pay. According to Liang, compared with other sectors, companies and professionals from the financial industry have a much bigger budget to pay for tech solutions like Tigerobo’s. "As a young startup, it’s important to sustain operations through generating revenue on our own besides receiving funds from investors,” she said.

Tigerobo’s capability of extracting useful information from unstructured data makes it especially appealing to clients in the financial sector, who usually have to deal with massive amounts of data, for increased efficiency and lowered labor costs.

The company monetizes Tigerobo Search through a subscription-based business model instead of advertisements. More individuals, including investors and journalists, subscribe to Tigerobo Search than businesses and the renewal rate is high. “Premium users account for a large portion of Tigerobo’s users and the repurchase rate is more than 40% [during the Covid-19 pandemic],” said Chen. 

B2B modular solutions

Tigerobo made a tactical switch while selling its flagship product, Tigerobo Search, which was meant to help business clients increase efficiency but was attracting more non-business, individual customers. 

“We soon found out that professional institutions, like securities companies, had already been using professional software to monitor the market, for example, Wind Financial Terminal. It was not easy to convince them to switch to another tool,” said Liang.

Nevertheless, Tigerobo continues to prioritize B2B customers, where most of its revenue comes from. According to Chen, it is natural for an AI company to start with serving business client and then shift its focus to individual users. 

Tigerobo now sells modular solutions to its business clients based on the tech used to create Tigerobo Search. These solutions can be incorporated into portals or information applications of clients to provide smart search service, algorithm support, intelligent chatbot and public opinion monitoring service. A securities company saw a 60% increase in traffic after integrating Tigerobo’s B2B solutions into its own portal.

Meanwhile, the company’s clientele is expanding beyond the finance industry. Government departments as well as businesses in the energy and media sectors are also adopting Tigerobo’s products, with per customer transaction amounting to about the 7-digit RMB mark. “An energy company has used our solution to filter SMEs’ credit records for its financial business,” said Liang. 

In the short term, B2B is the company’s focus. “We are seeing a clear trend that companies from less-digitized industries are rushing to work with startups that can provide advanced technology to help them digitize,” said Chen.

“In the long run, our goal is to launch a wide range of information products targeting individual users. The experience of working with companies in various industries will help us make such products more powerful and disruptive.”

In August 2019, the company applied its NLP-based search technology to an intelligent chatbot that answers users’ questions on WeChat. The company plans to combine the chatbot with smart speakers next. 

“You can acquire knowledge and information just by talking to it [the chatbot] in plain language,” said Chen.

Edited by S. Mani, Wang Xiao'e

FEEDBACK / COMMUNITY CONTRIBUTIONS

We pride ourselves on the accuracy of our information and reporting. Please help us by letting us know of any incomplete or inaccurate information on our website.

You’re reading Premium content, now open for limited access

Your payment was not successful.

Please make sure you have entered your payment details correctly. Or try again in a few moments.

small logo

The discount code you entered is invalid

Please make sure you have entered your discount code correctly. Or try again in a few moments.

Download successful.

Your sample has been sent. Please check your email.

By continuing to browse this website, you consent to our use of cookies, as well as to our Terms of Use and Privacy Policy which provide additional information about how we process your data. This website uses cookies to enhance your user experience. Please read our Cookies Policy for more information on how we use cookies, as well as instructions on how to disable cookies. You may disable cookies through your internet browser settings, however this may result in some parts of the website not working properly for you.