dineExpand

Dining-specific semantic dictionary expander for hospitality research

dineExpand expands seed dictionaries for dining and hospitality research using word embeddings pre-trained on 512,140 Yelp restaurant reviews. Any theoretical construct — revisit intention, food neophobia, authenticity, service quality, price perception — can be expanded with a single function call and restaurant-domain vocabulary out of the box.

Built on lexiExpand for the core similarity engine. Use dineExpand when your research is dining-specific; use lexiExpand for any other domain.

---

Installation

# install.packages("remotes")
remotes::install_github("gvajpai/dineExpand")

---

Quick start

library(dineExpand)

# Step 1 — Load restaurant vectors (~33 MB, one-time download, then cached)
vecs <- load_dining_vectors()

# Step 2 — Expand any dining construct
expand_dining(c("revisit", "return", "loyalty"), vectors = vecs)

# Step 3 — Export
lexiExpand::export_dict(result, name = "revisit_intention", format = "list")

---

Usage examples

Example 1 — Single seed word

vecs <- load_dining_vectors()

expand_dining("aroma", vectors = vecs, interactive = FALSE)
#>        word similarity  seed pct_match
#> 1     scent      0.841 aroma       84%
#> 2  smokiness      0.823 aroma       82%
#> 3   fragrant      0.809 aroma       81%
#> 4    flavors      0.796 aroma       80%
#> 5     savory      0.781 aroma       78%

---

Example 2 — Vector of seeds (hospitality constructs)

# Revisit intention
expand_dining(
  seed        = c("revisit", "return", "loyalty"),
  n           = 15,
  threshold   = 0.65,
  vectors     = vecs,
  interactive = FALSE
)

# Food neophobia — centroid mode finds words near the concept centre
expand_dining(
  seed        = c("adventurous", "novel", "unfamiliar", "exotic"),
  seed_mode   = "centroid",
  vectors     = vecs,
  interactive = FALSE
)

# Authenticity
expand_dining(
  seed        = c("authentic", "traditional", "genuine", "local"),
  vectors     = vecs,
  interactive = FALSE
)

# Service quality
expand_dining(
  seed        = c("attentive", "courteous", "prompt", "helpful"),
  vectors     = vecs,
  interactive = FALSE
)

---

Example 3 — Custom lexicon data frame (word + dimension)

expand_dining_lexicon() accepts any two-column data frame and expands every dimension in one call.

my_dict <- data.frame(
  word      = c("aroma",     "flavor",    "texture",
                "delighted", "excited",   "moved",
                "revisit",   "return",    "loyal",
                "adventurous","novel",    "unfamiliar"),
  dimension = c("sensory",   "sensory",   "sensory",
                "affect",    "affect",    "affect",
                "revisit",   "revisit",   "revisit",
                "neophobia", "neophobia", "neophobia")
)

candidates <- expand_dining_lexicon(
  my_dict,
  n         = 15,
  threshold = 0.65,
  vectors   = vecs
)

# Results include a $dimension column
table(candidates$dimension)
#>     affect  neophobia    revisit    sensory
#>         14         12         13         17

# Export as quanteda dictionary
library(quanteda)
lexiExpand::export_dict(
  candidates,
  seed   = my_dict$word,
  name   = "dining_constructs",
  format = "quanteda"
)

---

Train on your own corpus

Supply your own restaurant reviews to train domain-specific vectors:

library(data.table)
reviews <- fread("my_reviews.csv")

vecs_custom <- train_dining_vectors(
  texts      = reviews$text,
  dims       = 100L,
  cache_path = "~/my_restaurant_vectors.rds"
)

expand_dining(c("aroma", "flavor"), vectors = vecs_custom)

---

How the default vectors were trained

Property	Value
Corpus	Yelp Open Dataset — restaurant reviews
Reviews	512,140
Algorithm	GloVe (Pennington et al., 2014)
Dimensions	100
Window size	5
Iterations	20
Vocabulary	~98,000 terms
File size	33.2 MB

Restaurant-specific vocabulary clusters meaningfully in this space. For example, attentive → waitstaff, knowledgeable, courteous, prompt; and aroma → smokiness, savoriness, fragrant — terms that would rank far lower in Wikipedia-trained embeddings.

---

Function reference

Function	Description
load_dining_vectors()	Download & cache Yelp-trained vectors (run once)
expand_dining()	Expand a seed word or vector of seeds
expand_dining_lexicon()	Expand a full lexicon data frame by dimension
train_dining_vectors()	Train custom vectors from your own corpus
lexiExpand::export_dict()	Export to list, data.frame, or quanteda dictionary

---

Related packages

Package	Purpose
lexiExpand	General-purpose semantic expander (any domain)
mdeinR	Memorable Dining Experience dictionary and scoring

---

Citation

If you use dineExpand in published research, please cite:

Vajpai, G. N. (2026). dineExpand: Dining-Specific Semantic Dictionary Expander for Hospitality Research. R package version 0.1.0. https://github.com/gvajpai/dineExpand

---

License

MIT © 2026 Gopi Nath Vajpai