examplecode

## Example: Classifying a small text dataset

So far, we have only sent a single paragraph to the API. In most real projects, however, you will want to classify or transform many texts at once (for example survey answers, tweets, or news leads).

Below, we use the small play dataset data/play_data_climate.csv (100 short sentences about climate policy) and classify each text into three stance categories: Pro, Neutral, or Contra.

### 1. Load the example dataset

We first load the CSV file and inspect the first rows.

```{r eval=FALSE}
library(readr)
library(dplyr)

#read the play dataset from the repository
play_data <- read_csv("data/play_data_climate.csv")

#inspect the first rows
play_data %>%
slice_head(n = 5)
```

### 2. Define a prompt template for stance classification

Next, we define a helper function that builds a prompt for each text.
The model is instructed to return only a JSON object with a single field label.

```{r eval=FALSE}
build_prompt <- function(text) {
paste(
"You are a classifier for stance towards climate policy.",
"Classes: Pro, Neutral, Contra.",
"Return ONLY a valid JSON object with the field 'label'.",
"",
"Text:", text,
sep = "\n"
)
}
```

### 3. Send classification requests to the OpenAI API

We now create a helper function that:

- Builds the prompt for a given text.
- Sends the request to the OpenAI chat completions endpoint.
- Parses the raw JSON response into an R list (still not “nice” yet).

This code assumes that you have already set:
api_key
base_url <- "https://api.openai.com/v1/chat/completions"
headers (including the Authorization and Content-Type headers)

```{r eval=FALSE}
library(httr)
library(jsonlite)

classify_text <- function(text, api_key) {
prompt <- build_prompt(text)

body <- list(
model = "gpt-3.5-turbo",
messages = list(
list(role = "user", content = prompt)
),
max_tokens = 50,
temperature = 0
)

res <- httr::POST(
url = base_url,
body = jsonlite::toJSON(body, auto_unbox = TRUE),
encode = "raw",
headers
)

httr::stop_for_status(res)
jsonlite::fromJSON(
httr::content(res, as = "text", encoding = "UTF-8")
)
}
```

### 4. Classify a small subset of the data

In a real project, you might loop over thousands of texts.
For this tutorial, we only classify the first few rows to keep runtime and costs low.

```{r eval=FALSE}
sample_data <- play_data %>%
slice_head(n = 5)

#run the classification for each text in the sample
results_raw <- lapply(sample_data$text, classify_text, api_key = api_key)

#look at the first raw response
results_raw[[1]]
```

In the next step, you can show how to turn these raw JSON responses into a clean data frame (for example by extracting the label field for each text and joining it back to sample_data).