How Online Doctor Reviews Affect Online Consultation Demand

Online doctor reviews are often seen as digital word-of-mouth: helpful, messy, and hard to compare. A five-star average might look reassuring, but it doesn’t tell you what patients actually experienced — whether they felt listened to, whether the visit solved the problem, or whether they waited too long for an answer.

This matters in telehealth because patients often have to choose a doctor with limited information. They can’t “test” the experience ahead of time. The platform page becomes the storefront, and reviews become one of the few ways to judge what it might feel like to get care from a particular clinician.

A paper in the Information Systems Journal by Bin Zhang, an associate professor in the Department of Information and Operations Management at Texas A&M University’s Mays Business School, together with Haijing Hao, Yongcheng Zhan, and Jiang Wu, takes a practical step: it turns unstructured review text from online physician reviews into a consistent set of service-quality signals for each doctor and then tests whether those signals line up with weekly online consultation demand.

The core question is simple: do the service-quality signals inside online doctor reviews help predict which doctors get more online consultations, and can a healthcare-focused small language model extract those signals reliably? In this dataset, the authors find higher service-quality scores derived from review text are associated with higher online consultation demand.

Why “Service Quality” in Reviews Is Hard To Measure, and Why It Matters

Review text is detailed, but not organized. People write in their own style, with their own priorities. One person may focus on bedside manner. Another may focus on whether the doctor’s advice worked. A third may mention scheduling delays or communication speed.

Many tools that analyze reviews collapse all that nuance into something generic, like “positive vs. negative.” The authors argue approach can miss what matters in healthcare, because healthcare quality isn’t a single trait. Patients often evaluate multiple parts of the experience at once, including safety concerns, perceived effectiveness, how respected they felt, and how smoothly the process worked.

The challenge, then, is consistency. If you want to learn from reviews at scale, you need a way to translate millions of sentences into signals that mean the same thing across doctors and time. Otherwise, it’s hard to separate meaningful patterns from noise.

The Dataset: A Large Chinese Platform and Millions of Reviews

The study uses data from HaoDF.com, described as one of China’s largest platforms for online doctor reviews and consultations. The authors collected 24 months of data (February 2022 to February 2024) via web crawling.

They focus on online consultation reviews and narrow the sample to doctors treating selected low-risk and high-risk diseases. The resulting dataset includes:

• 5,258 doctors
• 1,316,771 online doctor reviews

To measure demand, the authors use the number of weekly online consultation appointments per doctor. They also track other visible platform signals, including review counts, thank-you letters, gifts, numerical ratings, and a platform recommendation score.

That combination is important because it helps separate “what the platform shows” (like counts and ratings) from “what the review text contains” (specific experience signals).

A Healthcare Lens for What Patients Write About

Rather than letting an algorithm invent its own topics, the authors apply a healthcare quality framework and translate review text into scores for five specific service-quality areas:

• Safety: avoiding harm and errors
• Effectiveness: achieving desired outcomes
• Patient-centeredness: responsiveness and empathy
• Timeliness: minimizing delays and wait time
• Efficiency: avoiding waste and unnecessary steps

These categories are easy to skim, but they’re also practical. They map closely to the kinds of things patients bring up in everyday language. For example, patients may not use the word “patient-centeredness,” but they often describe feeling rushed, ignored, reassured, or supported. Likewise, they may not say “efficiency,” but they may describe repeated steps, unnecessary back-and-forth, or a process that felt smooth.

The paper’s goal is to make those themes measurable, so they can be compared across doctors, and linked to demand.

How the Analysis Works

The study’s logic can be understood as a simple pipeline:

1. Start with written reviews. The authors collect online consultation reviews posted on the platform.
2. Convert review text into service-quality scores. A healthcare-focused small language model reads the text and generates scores for the five dimensions (safety, effectiveness, patient-centeredness, timeliness, efficiency).
3. Aggregate scores by doctor and week. Instead of treating each review as a one-off, the study summarizes the service-quality signals a doctor is receiving in a given week.
4. Compare those scores to weekly demand. Demand is measured as the weekly number of online consultation appointments per doctor.
5. Account for other visible signals. The analysis also considers other platform information such as ratings, review counts, and recommendation score.

Readers don’t need the engineering details to follow the central idea: the model makes review text comparable, and the study tests whether those comparable measures are associated with demand.

What Doc-BERT Does

The paper introduces Doc-BERT, a smaller language model made for healthcare review text. Instead of using a general-purpose method, it reads reviews and produces sentiment-style scores aligned to the five service-quality areas. For most readers, the key point is not the architecture. It’s the function: the model turns free-form writing from online doctor and physician reviews into structured signals that can be tracked over time. That creates a bridge between “qualitative” feedback (review text) and “quantitative” outcomes (weekly booking demand).

The paper reports this approach is more accurate than many other methods. The benchmark details aren’t necessary to understand the managerial implication: the authors treat these scores as reliable enough to use for the demand analysis.

What the Study Finds About Reviews and Online Consultation Demand

Across the authors’ analyses, a clear pattern appears: higher service-quality scores from review text are linked to higher demand for online consultations.

When the model considers all five service-quality areas together, each is statistically significant. The paper highlights two dimensions in particular:

• Effectiveness is the strongest predictor in the main model.
• Patient-centeredness is the next strongest.

Two additional findings help interpret how reviews work on platforms:

1. Review content adds information beyond review volume. When the service-quality dimensions are included, the relationship between the number of reviews and demand gets smaller. In other words, the text appears to explain something “counting reviews” can’t.
2. Recency matters for near-term demand. The paper finds recent reviews have a stronger relationship with consultation demand than older ones. This fits how people typically use reviews: a newer comment often feels more relevant than one from a year ago.

What This Means for Healthcare Managers and Platform Leaders

The study’s implications are practical. If demand aligns with measurable parts of service quality expressed in reviews, then review text can be more than reputation management. It can be a source of operational feedback, but only if it’s organized in a way leaders can act on.

Based on the paper’s results, two areas stand out as most aligned with demand in this setting:

• Effectiveness: whether patients perceive the care works
• Patient-centeredness: whether patients feel heard, supported, and responded to

For healthcare organizations, a sensible next step is to treat review text as a signal system, not just a comment stream. For example, leaders could:

• Monitor service-quality dimensions over time (weekly or monthly), not just overall ratings.
• Look for gaps between star ratings and dimension signals (e.g., high ratings but weakening timeliness comments).
• Use dimension trends to prioritize experience improvements that show up in patients’ language.

For platforms, the product implication is straightforward: star ratings alone may not be enough. Review text can carry richer information, especially when it is summarized into categories patients intuitively care about.

The paper does not specify policy implications beyond discussing how such measurement could support evaluation and decision-making in online healthcare settings.

What the Study Does Not Claim (and Why That Matters)

This study is an observational analysis of a single platform. Even with multiple modeling approaches and robustness checks, the paper does not claim improving a specific service-quality area will directly cause an increase in demand.

The paper also notes several limitations. It does not include offline clinical outcomes for each doctor, lacks detailed patient-level data that could affect selection, and does not fully track offline consultations. The authors also warn that differences across countries and health systems may affect how broadly the findings apply.

“How Physician Reviews Affect Online Consultation Demand: An Innovative Small Language Model with Fine-Turning” is published in Information Systems Research Journal.

Featured Researcher

Bin Zhang

Associate Professor Department of Information and Operations Management

View Profile

Share this News