Package {llmjoin}

Type: Package
Title: LLM-Powered Fuzzy Join
Version: 0.3.0
Description: Resolves ambiguous links between data.frames using large language models (LLMs). Supports matching across spelling variations, translations, and differing levels of precision.
License: MIT + file LICENSE
Encoding: UTF-8
Config/testthat/edition: 3
Imports: httr, jsonlite, config, readr
Suggests: testthat (≥ 3.0.0)
Depends: R (≥ 4.2.0)
URL: https://github.com/evanliu3594/llmjoin
BugReports: https://github.com/evanliu3594/llmjoin/issues
Config/roxygen2/version: 8.0.0
NeedsCompilation: no
Packaged: 2026-06-08 18:22:33 UTC; Evan
Author: Yifan LIU [aut, cre]
Maintainer: Yifan LIU <yifan.liu@smail.nju.edu.cn>
Repository: CRAN
Date/Publication: 2026-06-16 19:50:02 UTC

Build a fuzzy-join joint data.frame via LLM

Description

Build a fuzzy-join joint data.frame via LLM

Usage

build_joint(x, y, key1, key2, ...)

Arguments

x

a data.frame to be joined on the lhs.

y

a data.frame to be joined on the rhs.

key1

string, name of the key column of data.frame x waiting for pairing.

key2

string, name of the key column of data.frame y waiting for pairing.

...

extra params passed to chat_llm()

Value

a 2-column data.frame mapping values from key1 to key2.

Examples


  build_joint(
    x = data.frame(x = c("01","02","04")),
    y = data.frame(y = c("January","Feb","May")),
    key1 = "x", key2 = "y"
  )

Send message to LLM server

Description

This function sends a message to the LLM model and retrieves the result.

Usage

chat_llm(
  .message,
  .model = NULL,
  .temperature = 0,
  .max_tokens = 30000,
  .timeout = 300,
  .verbose = getOption("llmjoin.verbose", FALSE)
)

Arguments

.message

the message to send.

.model

character, LLM model to use. By default NULL (uses config value).

.temperature

OpenAI style randomness control (0~1), by default 0.

.max_tokens

Max tokens to spend.

.timeout

Max seconds to communicate with LLM.

.verbose

logical, print progress messages. Default getOption("llmjoin.verbose", FALSE).

Value

A character string with the LLM's response text.

Examples


  chat_llm("tell a joke.")

Generate connector prompt

Description

Generate a prompt to guide the LLM in generating a joint for data frame joining, leveraging the two key columns from the tables to be connected. As of 2025/04/10, DeepSeek R1 and gpt-4.1-mini showed the best result; other LLMs might fabricate non-existent data in the result.

Usage

joint_prompt(x, y)

Arguments

x

1-column data.frame or vector of characters, left hand side of the join

y

1-column data.frame or vector of characters, right hand side of the join

Value

A character string containing the matching prompt.

Examples

joint_prompt(
  data.frame(x = c("01","02","04")),
  data.frame(y = c("January","Feb","May"))
)

Fuzzy join with LLM

Description

Fuzzy join with LLM

Usage

llm_join(x, y, key1, key2, ...)

Arguments

x

a data.frame to be joined on the lhs.

y

a data.frame to be joined on the rhs.

key1

string, name of the key column of data.frame x waiting for pairing.

key2

string, name of the key column of data.frame y waiting for pairing.

...

extra params passed to chat_llm()

Value

the fuzzy-joined data.frame

Examples


  x <- data.frame(id = c("01", "02", "04"), value = c(10, 20, 40))
  y <- data.frame(month = c("January", "Feb", "May"), amount = c(100, 200, 400))

  llm_join(x, y, key1 = "id", key2 = "month")

Parse LLM response into a fuzzy-join joint data.frame

Description

Strips markdown fences, extracts the longest consecutive block of comma-separated lines, ensures a header row matching 'key1,key2' is present, and parses the CSV into a 2-column data.frame.

Usage

parse_joint(llm_response, key1, key2)

Arguments

llm_response

character, raw response from the LLM.

key1

string, name of the lhs key column.

key2

string, name of the rhs key column.

Value

a 2-column data.frame mapping values from key1 to key2.

Examples

parse_joint("01,January\n02,Feb\n04,May", key1 = "id", key2 = "month")

Set up your LLM service

Description

Set up your LLM service with native support for OpenAI, Claude (Anthropic), and Gemini (via OpenAI-compatible endpoint). For custom endpoints like Ollama, proxies, DeepSeek, Kimi, and others, use provider = "openai" along with your custom URL to connect through the compactible API interface. All information is stored strictly locally in your system configuration and is never uploaded or shared.

Usage

set_llm(provider = "openai", url = NULL, key = NULL, model = NULL)

Arguments

provider

character, LLM provider. One of "openai", "claude", "gemini". Default "openai".

url

url to your LLM provider endpoint. If NULL, auto-set based on provider.

key

api-key of your service.

model

character, model name. If NULL, auto-set from provider default.

Value

NULL invisibly. Called for side effect of writing the config file.

Examples


  set_llm(provider = "openai", key = "<your-openai-api-key>", model = "gpt-5.4-mini")

Convert a data frame to a markdown table

Description

Convert a data frame to a markdown table

Usage

tbl2md(tbl, nm = NULL)

Arguments

tbl

a data.frame object or a vector.

nm

character, only used if 'tbl' is a vector.

Value

markdown style table string lines

Examples

tbl2md(iris)