Arxiv Citation Recommendations via collaborative filtering

ArxivCitationRecommender.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "ArxivCitationRecommender.ipynb",
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FgiPvjubRjfz",
        "colab_type": "text"
      },
      "source": [
        "## Citation Recommendations via Collaborative Filtering\n",
        "\n",
        "Grab the [internal citation data](https://www.kaggle.com/Cornell-University/arxiv?select=internal-citations.json) from the Kaggle dataset and upload it here.\n",
        "\n",
        "*Thanks to [this fastai.collab tutorial](https://towardsdatascience.com/collaborative-filtering-using-fastai-a2ec5a2a4049)*"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YQLo0E9TRtgA",
        "colab_type": "text"
      },
      "source": [
        "### Imports"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "A9BP2qm9LHWX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import fastai\n",
        "from google.colab import files\n",
        "import json\n",
        "import pandas as pd\n",
        "from tqdm import tqdm"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vLeyIORDRzbG",
        "colab_type": "text"
      },
      "source": [
        "### Upload/unzip citation data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hblNi-B1LJRw",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "files.upload()"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qAQjUy4-LUXl",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        },
        "outputId": "4a60f36a-f9bf-4e4c-ae59-81da03acae0b"
      },
      "source": [
        "!unzip 612177_1135627_compressed_internal-citations.json.zip"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Archive:  612177_1135627_compressed_internal-citations.json.zip\n",
            "replace internal-citations.json? [y]es, [n]o, [A]ll, [N]one, [r]ename: "
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UyXB-TztR67B",
        "colab_type": "text"
      },
      "source": [
        "### Get citation JSON"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "q0fxHmpoLhG3",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "with open(\"internal-citations.json\", \"r\") as fp:\n",
        "  citations = json.loads(fp.read())\n",
        "\n",
        "print(list(citations.keys())[:10])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "z-bWdHr9MHAB",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "for paper, list_of_citations in list(citations.items())[:10]:\n",
        "  print(\"{}: {}\".format(paper, list_of_citations))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "M0zzxz1VR9ql",
        "colab_type": "text"
      },
      "source": [
        "### Generate Citation DF"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "6P6Vo0mqMQYF",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# generate CSV\n",
        "citing_papers = []\n",
        "citees = []\n",
        "scores = []\n",
        "for paper, list_of_citations in tqdm(citations.items()):\n",
        "  for citation in list_of_citations:\n",
        "    citing_papers.append(paper)\n",
        "    citees.append(citation)\n",
        "    scores.append(1.0)\n",
        "\n",
        "citation_df = pd.DataFrame({'paperID': citing_papers, 'citationID': citees, 'target': 1.0})"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3gORPN8aNeTk",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "citation_df"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gRGwYtRmSDGj",
        "colab_type": "text"
      },
      "source": [
        "### CF via fastai"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VnSnzTqqNnd_",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from fastai.collab import *\n",
        "from fastai.tabular import *"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ESwlE3vJNv-W",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "data = CollabDataBunch.from_df(citation_df, seed=42, valid_pct=0.2)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ylhKS5PAN39l",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "y_range = [0.0, 1.0]\n",
        "learn = collab_learner(data, n_factors=50, y_range=y_range, wd=1e-1)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8uUDFUfZOHHy",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "learn.fit_one_cycle(5, 5e-3)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7ZnyuIYiOPRx",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}