mdeinR

Memorable Dining Experience (MDE) text analysis for restaurant reviews

mdeinR scores the five dimensions of the Memorable Dining Experience (MDE) framework from free-text restaurant reviews using a theoretically grounded, validated lexicon of 324 words.

Dimension	Description	Words
sensory	Sight, sound, taste, smell, touch; food quality, ambience, décor	85
affect	Emotions and psychological reactions aroused during dining	82
social	Feelings of belonging; interactions with staff, friends, family	58
intellectual	Curiosity, learning, cognitive engagement	51
behavioral	Physical engagement, activities, involvement	48

The package implements sentence-level MDE scoring with valence-shifter adjustment (negator / amplifier / de-amplifier / adversative conjunction), a built-in sentence tokeniser, and a bundled valence-shifter table — all without external NLP dependencies. The scoring algorithm is inspired by sentimentr by Tyler W. Rinker.

---

Installation

# Install devtools if needed
# install.packages("devtools")

devtools::install_github("gvajpai/mdeinR")

After cloning or installing, build the data objects once:

source("data-raw/build_dictionary.R")

Dependencies

mdeinR only requires:

Package	Role
data.table	Fast in-memory operations
stringi	String tokenisation

No external NLP package is required.

---

Quick start

library(mdeinR)

# Built-in sample dataset — 400 reviews across 8 restaurants
head(restaurant_reviews)

# Sentence-level scores
mde(restaurant_reviews$text)

# Aggregate by restaurant
mde_by(restaurant_reviews, by = "restaurant")

# Aggregate by star rating
mde_by(restaurant_reviews, by = "stars")

# Annotate keywords in a single review
highlight_mde("The aroma was wonderful and staff were so caring.")

---

Valence shifters

mdeinR handles four types of valence shifters in a context window around each matched keyword:

# Negation — score clamped to zero
# 'ambiance' and 'beautiful' are both sensory words
mde(c(
  "The ambiance was beautiful.",
  "The ambiance was not beautiful."
))

# Amplification — score boosted by amplifier.weight (default 0.8)
# 'absolutely' amplifies 'stunning'
mde(c(
  "The aroma was stunning.",
  "The aroma was absolutely stunning."
))

# De-amplification — score reduced by n.neutral (default 0.2)
# 'somewhat' de-amplifies 'stunning'
mde(c(
  "The aroma was stunning.",
  "The aroma was somewhat stunning."
))

# Adversative — 'caring' after 'but' receives a reduced social score
mde(c(
  "The aroma was stunning and the hospitality was caring.",
  "The aroma was stunning but the hospitality was caring."
))

---

Algorithm

For each sentence:

1. Tokenise to lower-case words using Unicode-aware regex (punctuation stripped correctly).
2. For each token matching the MDE dictionary, assign raw score 1 / n to the matched dimension.
3. Examine a context window (n.before = 5, n.after = 2) for valence shifters from mdeinR::valence_shifters:
   • Negator (type 1): multiply score by −1 → clamped to 0
   • Amplifier (type 2): multiply by 1 + amplifier.weight (default 0.8)
   • De-amplifier (type 3): multiply by 1 − n.neutral (default 0.2)
   • Adversative (type 4): keyword after conjunction receives reduced weight
4. Per MDE theory, negated scores are set to zero — only positive occurrences count.
5. Dimension scores are summed across the sentence.

mde_by() averages sentence-level scores within each group.

---

Citation

If you use mdeinR in published research, please cite both the paper and the software:

Paper (lexicon and methodology):

Vajpai, G. N., Webb, T., & Beldona, S. (2025). Designing a memorable dining experience lexicon based on theory and text mining. International Journal of Hospitality Management, 130, 104245. https://doi.org/10.1016/j.ijhm.2025.104245

Software:

Vajpai, G. N., Webb, T., & Beldona, S. (2026). mdeinR: Memorable Dining Experience Text Analysis. Zenodo. https://doi.org/10.5281/zenodo.20215048

BibTeX:

@article{vajpai2025mde,
  title   = {Designing a memorable dining experience lexicon based on
             theory and text mining},
  author  = {Vajpai, Gopi Nath and Webb, Timothy and Beldona, Srikanth},
  journal = {International Journal of Hospitality Management},
  volume  = {130},
  pages   = {104245},
  year    = {2025},
  doi     = {10.1016/j.ijhm.2025.104245}
}

@software{vajpai2026mdeinR,
  title   = {{mdeinR}: Memorable Dining Experience Text Analysis},
  author  = {Vajpai, Gopi Nath and Webb, Timothy and Beldona, Srikanth},
  year    = {2026},
  doi     = {10.5281/zenodo.20215048},
  url     = {https://github.com/gvajpai/mdeinR}
}

---

License

MIT © 2025 Gopi Nath Vajpai, Timothy Webb, Srikanth Beldona

Portions of this package — specifically the sentence-level scoring algorithm, the valence-shifter parameter design, and the valence-shifter token lists — are derived from sentimentr by Tyler W. Rinker (MIT © 2017). Full derived-work notices are in LICENSE.md.