AlphaStream Personalization and Suggestion Systems 

Introduction 

AlphaStream is a leader in building personalized financial experiences for the financial markets. We work with regulated financial institutions like brokers and banks. We have a regulatory responsibility to explain how our models work as part of deploying AI into the regulated financial services sector. Still, we also get asked how the system works and how do you create a personalized financial experience for each user.

 

Personalization and Suggestion systems 

Let’s delve into AlphaStream’s personalization and suggestion system.

Our Director of Product describes: “The Financial Experience Cloud and the models that operate within it are seeking to build an increasingly personalized view of the financial markets, surfacing only relevant content and data and suggesting instruments for research that best match the financial market preferences of each user that reflect active market conditions.”

In essence, we cut through the noise, highlight what is important to the user, show market opportunities and risks and create a highly engaging financial experience unique to every user.

To do this, the Financial Experience Cloud must capture and process data on three key behavior verticals:

    • Financial Markets (Price, performance, Economics) – Technical and Fundamental data regarding financial instruments such as price action or earnings results.
    • Financial Content (News, Commentary, Analysis) – Processing and Analysis of what the market is talking about across multiple public and private sources
    • User and Community (Individual and Global Activity) – Trading and interaction data from all users across all platforms and apps within the financial institution

For more information on the 3 behavior verticals, please read it here. 

The AlphaStream platform implements a model with two functions, personalization (relevant, accurate, and real-time) and suggestion (related, active, and engaging).

The personalization model selects financial content and data that best matches the financial market preferences of the individual. These preferences are developed by weighting every user interaction on the platform and app by intent to research and trade; this generates a dynamic, accurate, and ranked list of instruments for each individual.

The model uses this data to filter and select content from the total content inventory. As new content is ingested into the Financial Experience Cloud, it is enriched with metadata through natural language processing. It allows the platform to curate it in real-time and deliver it to the display component, such as a content feed.

The personalization model is the baseline financial experience generator that continually improves the relevance and accuracy of the experience. The suggestion model is about growth, introducing new preferences to the individual based on historical preferences, related instruments, and real-time activity in Financial Markets, Content, and User/Community behavior.

The suggestion model differs from other well-known suggesters, such as Spotify and Netflix, in that the content and financial instrument continuously change rather than being constant. For example, the traits of a movie or a song (actors, beat, genre) do not change, even if the user’s tastes do. Comparatively, characteristics of a financial instrument change in seconds; price, volume, sentiment and commentary, performance, management, and other global factors all fluctuate rapidly and are all vital characteristics for a trader, investor, or finance professional to understand a financial instrument properly and therefore market and individual taste.

The suggestion model surfaces instruments and content related to individual preferences and is generated from collaborative filtering of the global and cohort preferences. Suggestions play an important role in introducing new instruments into the financial experience and combat echo chamber effects that the personalization model could reinforce.

However, there is a challenge when building a suggested instruments list too many of them. A simple example is if a user has interacted with or traded Nvidia in the past, then the model will immediately suggest similar or familiar equities such as AMD or Intel, which the user will be more likely to be interested in than, say, BHP or Unilever. Suppose the user has more than, say 10 instruments in their preferences. In that case, this will generate more than a hundred related instruments alone, then add collaborative filtering suggestions which suggest instruments that people like you have found engaging. You quickly get an unmanageable list you can’t display to the individual. You need to prioritize those suggestions for each individual somehow.

This is where AlphaStream leverages the data generated from the 3 behavior verticals to score each instrument suggested to an individual. This scoring provides the method to explain the activity of an instrument in the financial markets, content, and user community at any given time.

AlphaStream uses the data generated in the 3 behavior verticals to score the activity level of each instrument at any given time. The activity scoring of an instrument is looking to assess if there is a trading opportunity or risk for the individual to research and that it should be prioritized over other instruments, which may need to be more engaging at that time.

Ranking suggested instruments by current activity across multiple behavior verticals enables hundreds of instruments to be reduced to a handful that are not only familiar but are highly engaging because of what’s happening in the markets, the news, or the trading community.

A simple example of how an active suggestion outperforms a static preference-related suggestion would be an equity with an earnings release tomorrow, which is of the same importance to the user as one that does not have an imminent earnings release. Instrument activity is a significant contributor to engagement.

The AlphaStream model creates two lists of instruments for each individual; one is used for personalization, which is a very accurate view of their financial market preferences and suggestions, which is the most engaging instrument for that individual at any given time and may become a financial preference itself.

This conversion from a suggestion to a preference is an essential mechanism in assessing the performance of our personalization and suggestion models.

AlphaStream doesn’t consider a suggestion successful until it has achieved a specific score as a financial preference. This means that a suggestion would typically need several interactions of different types and intent weightings to confirm success and therefore feed back to the model. A simple example would be ADM being suggested in a component such as ‘trends for you’ and included in a content feed. The user clicking on a link, clicking on a tweet, performing a search, and changing a chart, all interaction show intent and should confirm ADM as a good suggestion that increases engagement and may lead to a trade.

AlphaStream’s goal as a platform is to deliver a personalized and ever more relevant and engaging experience, improving trading performance for the customer and profitability for the financial institution. Our models that sit at the heart of the platform are only a part of delivering a real-time financial experience but one that we are passionate about improving as we continue this journey of building the Financial Experience Cloud.

Written by
Adam Howard

Might Interest You

Privacy and personalization on mobile apps

Read more

Benzinga Best New Product 2022 Award

Read more

Highlights of AlphaStream’s 2022: A Year in Review 

Read more

Privacy and personalization on mobile apps

Read more

Benzinga Best New Product 2022 Award

Read more

Highlights of AlphaStream’s 2022: A Year in Review 

Read more