Statistical Relational Learning Towards Modelling Social Media Users

Golnoosh FarnadiIJCAI 2015 · OpenReview

Abstract

Nowadays web users actively generate content on different social media platforms. The large number of users requiring personalized services creates a unique opportunity for researchers to explore user modelling. To distinguish users, recognizing their attributes such as personality, age and gender is essential. Traditional machine learning approaches often treat user attributes as isolated variables, failing to capture the complex interactions and dependencies inherent in social media. We introduce a new statistical relational learning (SRL) framework, $PSL^Q$, which incorporates soft quantifiers to better model social media users by accounting for nuanced social influences and noisy data.

Introduction

Modeling social media users is notoriously difficult because human behavior isn't binary, and users don't exist in a vacuum. While Statistical Relational Learning (SRL) attempts to model the connections between users, it typically relies on standard first-order logic quantifiers: Universal ($\forall$) and Existential ($\exists$). However, social influence rarely operates on these extremes. For instance, a user might be influenced if "most" or "a few" of their friends exhibit a certain behavior, rather than requiring all or just one. Standard SRL frameworks struggle to express these "soft" thresholds, limiting their ability to capture the true dynamics of social media.

Methods

The proposed framework, $PSL^Q$, extends Probabilistic Soft Logic (PSL) by incorporating "soft quantifiers" like "most" or "a few".

Continuous Truth Values and Soft Quantifiers

Unlike Boolean logic, PSL allows atoms to take values in the interval $[0, 1]$, which is ideal for representing the varying degrees of friendship or trait strength in social networks. $PSL^Q$ introduces a syntax for quantifiers $Q$ based on Zadeh’s approach to fuzzy logic. These quantifiers are defined by membership functions that map a proportion of a group to a truth value.

Membership functions for soft quantifiers "Few" and "Most".

As shown in Figure 1, "Most" might reach a truth value of 1.0 only when 75% of the target group meets a criteria, while "Few" reaches it much earlier (e.g., at 40% of the population).

Inference

Adding soft quantifiers introduces non-linearity into the optimization problem. To address this, the authors developed an iterative MPE (Most Probable Explanation) inference algorithm. This algorithm decomposes the non-linear problem into a sequence of convex optimization problems, allowing for efficient solving using standard PSL techniques.

Results

The framework was evaluated on the Epinions dataset, focusing on modeling social trust. The task involved predicting trust between users based on triad relations (e.g., "If A trusts a few people who in turn trust C, then A likely trusts C").

Performance comparison between standard PSL and $PSL^Q$.

As shown in Figure 2, $PSL^Q$ outperformed the baseline PSL model across all metrics, including Precision (PR+, PR-) and Area Under the Curve (AUC). These improvements were found to be statistically significant ($p < 0.05$), demonstrating that the inclusion of soft quantifiers provides a tangible advantage in modeling relational trust.

Discussion and Conclusion

The results indicate that increasing the linguistic expressivity of a model—allowing it to reason with terms like "most" or "some"—directly correlates with improved predictive accuracy in social contexts. By using the quantifier "a few" in trust rules, the model becomes more flexible and resilient to noise compared to rigid universal quantifiers.

Future work includes automatically learning the mappings for quantifiers from data rather than manual definition, and exploring structure learning to automatically generate $PSL^Q$ rules from raw social media data. While the iterative solver is effective, further optimization will be necessary to scale the framework for massive, real-time social networks.

Summary updated

Abstract

1 Motivation Nowadays web users actively generate content on different social media platforms. The large number of users requiring personalized services creates a unique opportunity for researchers to explore user modelling. To distinguish users, recognizing their attributes such as personality, age and gender is essential. To this end, substantial research has been done by utilizing user generated content to recognize user attributes by applying different classification or regression techniques. Among other things, we have inferred the personality traits of Facebook users based on their status updates using three different classifiers [Farnadi et al., 2013]. But as we concluded in [Farnadi et al., 2014b], user attributes are not isolated, as emotions expressed in Facebook status updates are related to age, gender and personality of the authors. Using multivariate regression or multi-target classification techniques is one approach to leverage these dependencies in the user attribute learning process. For example, to employ the dependencies between different personality traits, we applied five multivariate regression techniques to infer the personality of YouTube vloggers [Farnadi et al., 2014c]. The above mentioned techniques are powerful types of machine learning approaches, however they only partially model social media users. Users communicate directly as friends, or indirectly, by liking others content. These types of interaction between users are a valuable source but modelling them with the traditional machine learning approaches is challenging. Furthermore, user generated content in social media comes in different modalities such as visual and textual content, whereas different pieces of evidence might contradict each other. Moreover, in extracting features from each source, a reasonable amount of noise is expected. Hence, using the extracted features as a single feature space without considering features’ noise, dependencies or conflicts, reduces the quality of the learning process. To overcome these limitations, we introduce a new statistical relational learning (SRL) framework [Getoor and Taskar, 2007] suitable for modelling social media users, which we call PSL Q [Farnadi et al., 2014a].
BibTeX
@inproceedings{farnadi2015statistical,
  author = {Golnoosh Farnadi},
  title = {{Statistical Relational Learning Towards Modelling Social Media Users}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2015},
  url = {https://openreview.net/forum?id=ry-K2fM_WB},
}
Figure