If a fintech company that has text data at its disposal isn’t using it to employ natural language processing models – a branch of artificial intelligence that teaches machines to understand, analyze and generate human language – it’s missing out.

Natural language processing or NLP models can and should be employed regularly to evaluate a company’s internal and external textual material to understand customer and employee sentiments. It can also be used to identify important business themes or trends that the company needs to evaluate and integrate into its business strategy.

This is especially true with the emergence of generative artificial intelligence, which makes natural language processing capabilities more powerful than ever.

That’s the clear message from data scientist Sumedha Rai in an interview with Fintech Nexus and in presentations at two recent conferences in New York this spring: the AI ​​in Finance Summit and the MLConf 2024 gathering of artificial intelligence and machine learning.

However, these are just two of the results that companies can obtain from continuous text analysis using NLP models.

Rai adds that such NLP tools, used in conjunction with other machine learning and artificial intelligence solutions, can also be used to quickly summarize and translate documents, understand important tags in text data, personalize customer interactions, and catch fraudsters by detecting anomalies in their communications. .

Sumedha Rai, senior data scientist

Rai is a senior data scientist at a microinvesting firm in New York City, where she spends a lot of time analyzing user sentiment and themes, examining data to assist in investment decisions, and evaluating fraud modeling. She also conducts research at the Center for Data Science and other affiliated departments at New York University.

He notes that perhaps the most important benefit that comes from regularly analyzing text via NLP – aside from increased efficiency – is that “people (employees) will have a lot more time to think about creative things,” related to product development and to your business strategy. , which constitutes a clear competitive advantage.

Text relevant to NLP analysis or summary includes everything from customer feedback, posts, complaints, social media comments, emails and survey results, transaction data, website Corporate web and internal data, employee communications, claims calls, agent feedback, regulatory, compliance and legal. data.

The benefit of quarterly or ongoing evaluation of such texts via NLP, Rai says, is that fintech companies can more easily personalize services, build better chatbots, detect fraud, summarize and translate global regulatory and compliance documents, and gain better understanding employee satisfaction. levels.

One type of text analysis, which uses NLP for topic modeling, can be used to track the topics that are most important in the minds of your customers, including what they like or don’t like about a product, and is a which Rai believes may be underutilized by many fintech companies.

Using this technique, “fintech companies should look at all their problems and challenges and see how much signal they received for these problems in the form of text. They should therefore leverage NLP analysis of text data to help solve many of these problems,” says Rai.

NLP models that can help in this exercise include Latent Semantic Analysis (LSA), Latent Dirichlet Allocation (LDA), LDA2vec and BERTopic and its different variants, however, for fintech companies in particular, using FinBERT, a transformation model that has been specifically pre-trained on financial text, it is also a great choice.

Among these model choices, however, Rai is particularly partial to BERT models because they are bidirectional in design and capture context based on this bidirectionality.

“They (BERT models) also have contextual embeddings, which allow the models to understand a word by considering all other words around it and take into account the context for each occurrence of a given word,” Rai says.

He adds: “Additionally, we now have access to powerful word embeddings from GenAI models, some of which are free to download. However, BERT is a great choice for establishing a baseline when working with LLMs, particularly when working with financial texts.”

Rai also highlighted the importance of taking full advantage of Named Entity Recognition (NER), a subfield of NLP that concerns the labeling of text so that named entities – single words, sentences or sequences of words – can be easily categorized .

“NER is a very underutilized commodity technology but, in reality, it can be employed in multiple ways to better understand which entities customers are most interested in, allowing you to better personalize your communications with them,” Rai says.

Note that NER analysis gives us a way to extract all critical information much faster from a large body of text and can be used to flag risky interactions or anomalies that could indicate potential fraud. In this way, it plays a vital role in continuous sentiment analysis and text classification.

One particularly useful feature, Rai says, is NER’s ability to help “document compliance at a glance very quickly,” so you can quickly extract key information from long documents and review it later efficiently.

With the introduction of generative AI models, Rai says, fintech companies now have access to a powerful tool for text analysis where minimal coding is involved, when using the out-of-box solution directly. However, the trade-off may be in the level of accuracy that may be lost in using off-the-shelf Gen AI models versus fine-tuning a model for specific tasks.

“Generative AI models are pre-trained, and so for simple text analysis, a pre-trained model can often do the job,” Rai says, adding that with multiple generative AI models to choose from, he favors the ease of use of Chat GPT which continues to improve in terms of accuracy and also has easily accessible APIs for integrating GPT models into your code.

He also believes that Meta’s LLAMA models, LLAMA 3 in particular, are powerful and useful, and are free to use.

However, Rai warns that fintech companies must keep in mind that there are risks in using out-of-the-box generative AI models.

“No sensitive or customer data should be entered into these models. These are hosted systems and the data comes out of the local machines and goes to a server where the model resides”, says Rai, underlining that the interaction data can be analyzed by companies that create LLMs to improve the performance and reliability of their systems .

“Even if you use the enterprise version of these models, I would still make sure your data has been stripped of all personally identifiable information (PII) before being entered into a model or used to query the model,” Rai says .

Evaluating models for bias, discrimination, data security, data privacy, hallucinations, and respectful content creation is also critical, Rai says, and starts by looking at what kind of data you’re feeding into the model, making sure all classes, genders and geographies are represented and also by employing a diverse team of people to work on the models instead of just one person.

Increasingly, Rai says, some fintech companies are hiring red teams from outside their company to conduct a thorough assessment and to ensure that a company’s working models have been “unbalanced.” do not generate distorted results that could give rise to discriminatory practices.

One way to save time on generation AI that Rai particularly appreciated was to ask Chat GPT to create a logo, slogan and launch press release for a fantasy fintech company.

“The results were impressive,” Rai said, noting that, on an ongoing basis, Chat GPT continues to improve and amaze.

• 

  Katherine Heires is a business and technology journalist and founder of MediaKat llc. As a freelance journalist, he covers a range of topics including the growing business impact of artificial intelligence and machine learning developments and trends related to fintech startups, embedded banking, open banking, behavioral finance, cybersecurity and fraud prevention technology. Her reporting on financial and fintech topics has appeared in Businessweek Online, Institutional Investor, Risk Intelligence, Risk Management Magazine and Venture Capital Journal.

Source