Jochen Hartmann

Jochen Hartmann

Auto-generated profile, based on publicly available information

Jochen Hartmann

Technical University of Munich, Munchen, Germany

Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
12 September 2023 in Business & Information Systems Engineering
Verified
Journal Article
Open access
PDF file
SFStefan Feuerriegel
Stefan Feuerriegel
JHJochen Hartmann
Jochen Hartmann
CJChristian Janiesch
Christian Janiesch
PZPatrick Zschech
Patrick Zschech

5 March 2024 in Customer Needs and Solutions
Verified
Journal Article
Open access
PDF file
JKJan Ole Krugmann
Jan Ole Krugmann
JHJochen Hartmann
Jochen Hartmann
In the rapidly advancing age of Generative AI, Large Language Models (LLMs) such as ChatGPT stand at the forefront of disrupting marketing practice and research. This paper presents a comprehensive exploration of LLMs’ proficiency in sentiment analysis, a core task in marketing research for understanding consumer emotions, opinions, and perceptions. We benchmark the performance of three state-of-the-art LLMs, i.e., GPT-3.5, GPT-4, and Llama 2, against established, high-performing transfer learning models. Despite their zero-shot nature, our research reveals that LLMs can not only compete with but in some cases also surpass traditional transfer learning methods in terms of sentiment classification accuracy. We investigate the influence of textual data characteristics and analytical procedures on classification accuracy, shedding light on how data origin, text complexity, and prompting techniques impact LLM performance. We find that linguistic features such as the presence of lengthy, content-laden words improve classification performance, while other features such as single-sentence reviews and less structured social media text documents reduce performance. Further, we explore the explainability of sentiment classifications generated by LLMs. The findings indicate that LLMs, especially Llama 2, offer remarkable classification explanations, highlighting their advanced human-like reasoning capabilities. Collectively, this paper enriches the current understanding of sentiment analysis, providing valuable insights and guidance for the selection of suitable methods by marketing researchers and practitioners in the age of Generative AI.

9 August 2023 in Journal of Consumer Psychology
Verified
Journal Article
Open access
JHJochen Hartmann
Jochen Hartmann
ABAnouk Bergner
Anouk Bergner
CHChristian Hildebrand
Christian Hildebrand
Prior research revealed a striking heterogeneity of how consumers view smart objects, from seeing them as helpful partners to merely a useful tool. We draw on mind perception theory to assess whether the attribution of mental states to smart objects reveals differences in consumer–smart object relationships and device usage. We train a language model to unobtrusively predict mind perception in smart objects from consumer‐generated text. We provide a rich set of interpretable linguistic markers for mind perception, drawing on a diverse collection of text‐mining techniques, and demonstrate that greater mind perception is associated with expressing a more communal (vs. instrumental) relationship with the device and using it more expansively. We find converging evidence for these associations using over 20,000 real‐world customer reviews and also provide causal evidence that inducing a more communal (vs. instrumental) relationship with a smart object enhances mind perception and in turn increases the number of tasks consumers engage in with the device. These findings have important implications for the role of mind perception as a novel lens to study consumer–smart object relationships. We offer an easy‐to‐use web interface to access our language model using researchers own data or to fine‐tune the model to entirely new domains.

1 January 2023 in SSRN Electronic Journal
Verified
Preprint
Open access
PDF file
JHJochen Hartmann
Jochen Hartmann
YEYannick Exner
Yannick Exner
SDSamuel Domdey
Samuel Domdey
Capable of generating photorealistic images, generative AI can augment human creativity and disrupt the economics of marketing content production. While anecdotal evidence hints at high-quality visual outputs of generative AI, little is known about these novel models’ effectiveness across marketing contexts. This paper explores the disruptive potential of text-to-image diffusion models for visual marketing content generation, systematically evaluating the performance of AI-generated vs. human-made images across three important marketing dimensions: human perception (quality, realism), social media engagement (likes, comments), and click-through rates (CTRs) of banner ads. In three preregistered studies, we collect more than 17,000 human evaluations for over 1,500 synthetic images generated with 13 text-to-image diffusion models. We obtain converging evidence that synthetic, AIgenerated images can (i) significantly outperform human-made images regarding perceived quality and realism, (ii) obtain comparable engagement levels in a social media context, and (iii) achieve a +22% higher CTR in a randomized field experiment with more than 86,000 impressions. Our findings suggest that the paradigm shift brought about by generative AI can help firms produce marketing content not only faster and cheaper but also at human quality and effectiveness levels with important implications for firms, consumers, and policymakers.

9 June 2026 in Nature Human Behaviour
Verified
Journal Article
Closed access
SFStefan Feuerriegel
Stefan Feuerriegel
CBChristopher Barrie
Christopher Barrie
MCM. J. Crockett
M. J. Crockett
LGLaura Globig
Laura Globig
KMKillian L. McLoughlin
Killian L. McLoughlin
MRManoel Horta Ribeiro
Manoel Horta Ribeiro

19 November 2019 in SSRN Electronic Journal
Verified
Preprint
Open access
PDF file
CSChristian Siebert
Christian Siebert
JHJochen Hartmann
Jochen Hartmann
MHMark Heitmann
Mark Heitmann
CSChristina Schamp
Christina Schamp
Every managerial and societal problem entails human opinions. Recent societal dynamics have vividly illustrated how opinions can become contagious, shape collective values, and decide about the fate of governments, corporations, and the trajectories of entire economies. Coping with these dynamics necessitates a reliable understanding of individual and societal sentiments. The basis for this exists with today's wealth of readily available opinion sources (e.g., social media, online reviews, financial reports, news articles). Concurrently with this proliferation of unstructured data, research in natural language processing has provided a plethora of methods to automatically approximate the sentiment a writer expresses to a reader. While computer scientists have demonstrated what can be achieved under optimal conditions, an empirical synthesis of this evidence and robustness tests of these achievements are still missing. For a few selected applications, the most recent developments suggest language models can achieve high accuracy also for sentiment analysis. However, their capabilities and relative advantages compared to alternatives have not been assessed for a broader range of data sets and applications. With insufficient knowledge about what a new technology can do and how it performs relative to established alternatives, adoption often remains below potential. In this paper, we assess the accuracy of language models across a broad range of 15 diverse applications with 392,967 labeled text observations. We compare language models with their main alternatives in a comprehensive meta-analysis (196 publications with 888 experimental outcomes based on over 50 million labeled documents). This represents the largest assessment of sentiment analysis accuracy we are aware of. We find that understanding which sentiment analysis method is best matters in terms of attainable accuracy. On average, language model-based sentiment classification matches human perception for 91.48% of observations. This high level of performance is significantly better than all available alternatives. In spite of their popularity, lexicons are the worst performing alternative with 20.13% less accuracy. We find no evidence for boundary conditions. These findings make language models a promising standard for sentiment analysis that can automatically capture human opinions with an accuracy that is likely sufficient for any typical application.