- User Modeling Definition
- Why personalization matters?
- Machine Learning for User Modeling
- User Modeling Examples
- Recommendation & Search Engines
- Product Management
- Ad Tech / Digital Marketing
- E-commerce Shopping Assistants
- Healthcare Diagnosis & Treatment
- User Modeling Principles
- User Modeling Approaches
- Static User Modeling
- Dynamic User Modeling
- User Modeling via Stereotypes
- Highly Adaptive User Modeling
The emergence of big data and machine learning opened up many possibilities for making business operation more efficient and precise.
Given the fact that every piece of content or an application produces a lot of data, it seems obvious to use it in a beneficial manner both for the end users and for the company. Before any of that happens, you need to understand who are you targeting at.
Understanding your target audience and delivering relevant content to them is one of the strategic priorities of any business operation.
User modeling's role is to prepare a solid frame of reference for understanding your target audience.
User modeling is a part of data mining operation explicitly designed for an exploration of the target audience (based on specific characteristics) and understanding the distinct patterns of its behavior.
In a way, it is a direct result of human-computer interaction. This contributes to the furthering and more elaborate segmentation of the target audience. In turn, the resulting model prepares the system for the user-adapted interaction.
Personalization, whether it is website content or advertisements, provides better precision of the targeting. It brings together the company and the customer, helping the company to define what is relevant to the particular segment of users and which type of content will click with them.
User modeling uses descriptive algorithms to find patterns and group them to offer suggestions regarding the content for a specific audience.
Overall, it is the process that consists of the following stages:
- composing a detailed “portrait” of certain user segments out of available data;
- integrating it into the system so that it will be capable of reacting to the actions of certain audience segments accordingly.
User Model is the result of this process, and its purpose is to:
- Define user intentions
- Explore user background (who is, where did he come from)
- Explore the context of use, i.e., user intent (for example, to keep notes or count sheep)
- Describe user traits and preferences
There is also a related term “user profile.” The difference between the user model and the user profile is that the latter is associated with a single user. In addition to that, user profiles are often using user-specified information while the user model is using information observed from the user’s actions.
Due to a large amount of information required for processing, User Modeling applies Machine Learning algorithms to enable its operation.
User Modeling commonly uses these methods of machine learning:
- Supervised Learning Classification: the algorithm is trained on labeled data and then used to classify new samples. In the meantime, it also detects unseen instances or anomalies.
- Supervised Learning Regression: the algorithm is trained to estimate the relationship and values of the variable elements.
- Unsupervised Learning Clustering: the algorithm is trained on the go with the unlabeled set of data in which it finds patterns and groups them accordingly.
- Random Forest method applies multiple decision trees to split up the data set into the relevant segments.
Recommendation engines and social networks are the primary fields where user models form the backbone of the entire operation.
Usually, these systems are used to present a selection of some content and also suggestions to interact with something (friends, promoted content, etc.), and user models create a "backdrop" for these actions.
- Google search engine uses available search history and profile data to adjust search results;
- Pinterest uses models to enable retargeting and mix it with the personalized content;
- Amazon uses user and searches history and behavioral data to manage its inventory around the preferences of the user and compile “You might also like” section;
- Netflix uses behavioral data and user history to compile the “You might also like” and “People also watch” sections;
- Facebook and Twitter are using models to construct relevant newsfeed and offer reasonable friend suggestions;
In these cases, the amount of information used to form the model or profile may vary. However, it is important to note that recommendation services predominantly fuel their operation out of behavioral data AKA the one demonstrated while using a specific function.
Monitoring the product use is the most basic way of applying user models. In this case, there are several goals behind its use:
- Critical assessment of the product use (who, why, when);
- Finding the problematic points in product use and solving them (design flaws, glitches, etc.);
- Ad retargeting (in case if there is ad content present)
As a result, you can get a big picture of how your product is used and what elements of the product need to be improved to keep the users from bouncing off. On the other hand, this information can help you understand how to market the product and what points to focus on.
Clear and relevant picture of a target audience, its needs, and preferences is in the heart of ad tech operation. User models and profiles form the backbone for the retargeting process.
Based on the user segments, ad content is shuffled around to deliver the best fitting piece at the right time.
The models are used on both ends of an operation. At first, they are used to form the campaign, then the information regarding interactions with ad content is later sent back to adjust models or profiles according to the performance results.
The purpose of user models in eCommerce is to amplify and maximize customer experience. eCommerce user models are used to adjust every element of the service according to the preferences of the user (explicitly stated or through demonstrated behavior) so that the shopping session will result in a purchase with a higher probability.
Overall, user models are used for the following features:
- Inventory customization shows products relevant to the user’s search queries, past preferences and additional data (in case the user came from an affiliate link);
- “You might also like” features group certain product types that the user might be interested in;
- Ad Retargeting presents relevant ad content according to the behavioral, contextual, affiliate data;
- Navigation assistance shows relevant types of products ranged in a specific category (price, brand, etc.);
- Showcasing relevant product information.
User data combined with a database that retains information about various diseases and treatment results can be helpful in further exploration of the treatment possibilities.
In this case, the critical factor is personalization and adjustment of both databases.
There are two ways such a combination can be used:
- User profiling can be used to present personalized suggestions regarding diagnosis or treatment in correlation with the state of health and certain treatment limitations (due to allergies, etc.).
- The system can be used in scenario simulation and testing out various courses of treatment without putting the patient in danger.
The challenge comes with the possible inconsistency of the patient data due to further developments of diseases or insufficient testing.
The process of user modeling is relatively easy to describe in plain language. While operating on the website or using an application — users are producing certain kinds of information.
Here are some of the more common examples of it:
- Transactional data (which gateway, currency, date, time, etc.)
- Product use data (various actions, usage stats, etc.)
- Demonstrated Web Behavior (session time, content preferences);
- Customer-related input text data (registration, comments, chatbot interaction, etc.)
- Social Media activity (logins, shares, etc.)
This information about users comes from two sources:
- the user itself (for example, upon registration or signing in via specific account or filling the form);
- monitoring on-site/in-app user behavior (which pages are visited, what kind of content got clicked and so on).
After that, the gathered information is fed into the data mining algorithm, which in turn takes out of its specific insights, the nature of which can be predetermined or discovered during the mining operation. For example:
- Certain types of personal information (name, age, gender, job title)
- Specific contact details (email, phone number, address)
- User source (where the user comes from — search results, social media, affiliate links, etc.)
- Content preferences (specific types of content, long session time, social media share, etc.)
- Reaction to particular kinds of content (clicks on ads, transitions, bounces, etc.)
- Lifetime value of a specific user (how much of the benefit you can generate from the user)
All this information is subsequently compiled into a user model/profile.
The result of an operation is a user profile that can be subsequently integrated into the system and used in a variety of ways.
For example, this information can be integrated into the digital marketing operation to keep retargeting precise. On the other hand, product use information or web behavior can be used to improve functional or design features of the product or website.
There are four major approaches to composing user models.
Static user modeling is the approach where the information is gathered once and remains unmodified.
This approach includes any given set of data (personal and behavioral). Many services that don’t require customization or any other type of adaptation (for example, forum, image boards, blog platforms) use this type of user modeling.
Static user modeling is also used for machine learning algorithm training.
Unlike the static approach, in the dynamic user modeling approach, the information is gradually updated. It may include various sets of data with multiple groupings. As a result, the system is capable of adjusting the delivery on the go.
The dynamic approach is commonly used in recommendation engines and Ad Tech retargeting operation.
User modeling via stereotypes means a generalized version of static profiles. Instead of making specific accounts for each user, user data is collected into larger chunks, united by common characteristics.
Once there is new incoming information, the stereotypes can be updated accordingly. This type of user modeling is preferable when you do not require personal information for effective operation.
Also, this is very useful in the European Union due to the adoption of the General Data Protection Regulation (GDPR).
Highly-adaptive user modeling is the opposite of the stereotypical approach. This type represents an extreme form of customization that requires as much information as possible to provide the result as precise as possible.
Given the fact that personal data-related legislation is getting stricter and also growing privacy concerns — this type of profiling will likely wind down and be limited to behavior-based data, but there is a chance it will evolve into something else.
Knowing your audience is the key to a successful business operation.
No wonder — without it, the whole thing will be more of a going all guns blazing with your eyes closed hoping to hit something instead of building up the engagement of the audience and as a result generating a steady stream of conversions.
Therefore, choose the type of user modeling that fits your requirements and expectations, and implement it into your business. You win because you get more precise targeting and your customers win because they get what they want and you seem to read their minds :)