FindDuplicateEmailIncidents

Source

import multiprocessing

# set omp
import os

import demistomock as demisto  # noqa: F401
from CommonServerPython import *  # noqa: F401

from CommonServerUserPython import *

os.environ["OMP_NUM_THREADS"] = str(multiprocessing.cpu_count())  # noqa
demisto.debug(f"Set OMP_NUM_THREADS to {os.environ['OMP_NUM_THREADS']}")

import re
from email.utils import parseaddr
from urllib.parse import urlparse

import dateutil  # type: ignore
import pandas as pd
import tldextract
from bs4 import BeautifulSoup
from FormatURLApiModule import *  # noqa: E402
from numpy import dot
from numpy.linalg import norm
from sklearn.feature_extraction.text import CountVectorizer

no_fetch_extract = tldextract.TLDExtract(suffix_list_urls=[], cache_dir=None)
pd.options.mode.chained_assignment = None  # default='warn'

SIMILARITY_THRESHOLD = float(demisto.args().get("threshold", 0.97))
CLOSE_TO_SIMILAR_DISTANCE = 0.2

EMAIL_BODY_FIELD = "emailbody"
EMAIL_SUBJECT_FIELD = "emailsubject"
EMAIL_HTML_FIELD = "emailbodyhtml"
FROM_FIELD = "emailfrom"
FROM_DOMAIN_FIELD = "fromdomain"
PREPROCESSED_EMAIL_BODY = "preprocessedemailbody"
PREPROCESSED_EMAIL_SUBJECT = "preprocessedemailsubject"
MERGED_TEXT_FIELD = "mereged_text"
MIN_TEXT_LENGTH = 50
DEFAULT_ARGS = {
    "limit": "1000",
    "incidentTypes": "Phishing",
    "existingIncidentsLookback": "100 days ago",
}
FROM_POLICY_TEXT_ONLY = "TextOnly"
FROM_POLICY_EXACT = "Exact"
FROM_POLICY_DOMAIN = "Domain"

FROM_POLICY = FROM_POLICY_TEXT_ONLY
URL_REGEX = (
    r"(?:(?:https?|ftp|hxxps?):\/\/|www\[?\.\]?|ftp\[?\.\]?)(?:[-\w\d]+\[?\.\]?)+[-\w\d]+(?::\d+)?"
    r"(?:(?:\/|\?)[-\w\d+&@#\/%=~_$?!\-:,.\(\);]*[\w\d+&@#\/%=~_$\(\);])?"
)

IGNORE_INCIDENT_TYPE_VALUE = "None"


def get_existing_incidents(input_args, current_incident_type):
    global DEFAULT_ARGS
    demisto.debug("Entering get_existing_incidents")
    get_incidents_args = {}
    get_incidents_args["limit"] = input_args.get("limit", DEFAULT_ARGS["limit"])
    if "existingIncidentsLookback" in input_args:
        get_incidents_args["fromDate"] = input_args["existingIncidentsLookback"]
    elif "existingIncidentsLookback" in DEFAULT_ARGS:
        get_incidents_args["fromDate"] = DEFAULT_ARGS["existingIncidentsLookback"]
    status_scope = input_args.get("statusScope", "All")
    query_components = []
    if input_args.get("query"):
        query_components.append(input_args["query"])
    if status_scope == "ClosedOnly":
        query_components.append("status:closed")
    elif status_scope == "NonClosedOnly":
        query_components.append("-status:closed")
    elif status_scope == "All":
        pass
    else:
        return_error(f"Unsupported statusScope: {status_scope}")
    type_values = input_args.get("incidentTypes", current_incident_type)
    if type_values != IGNORE_INCIDENT_TYPE_VALUE:
        type_field = input_args.get("incidentTypeFieldName", "type")
        type_query = generate_incident_type_query_component(type_field, type_values)
        query_components.append(type_query)
    if len(query_components) > 0:
        get_incidents_args["query"] = " and ".join(f"({c})" for c in query_components)

    fields = [
        EMAIL_BODY_FIELD,
        EMAIL_SUBJECT_FIELD,
        EMAIL_HTML_FIELD,
        FROM_FIELD,
        FROM_DOMAIN_FIELD,
        "created",
        "id",
        "name",
        "status",
        "emailto",
        "emailcc",
        "emailbcc",
        "removedfromcampaigns",
    ]

    if "populateFields" in input_args and input_args["populateFields"] is not None:
        get_incidents_args["populateFields"] = ",".join([",".join(fields), input_args["populateFields"]])
    else:
        get_incidents_args["populateFields"] = ",".join(fields)

    demisto.debug(f"Calling GetIncidentsByQuery with {get_incidents_args=}")
    incidents_query_res = demisto.executeCommand("GetIncidentsByQuery", get_incidents_args)
    if is_error(incidents_query_res):
        return_error(get_error(incidents_query_res))
    incidents_query_contents = "{}"

    for res in incidents_query_res:
        if res["Contents"]:
            incidents_query_contents = res["Contents"]
    incidents = json.loads(incidents_query_contents)
    demisto.info(f"Retrieved {len(incidents)} existing incidents for comparison.")
    return incidents


def generate_incident_type_query_component(type_field_arg, type_values_arg):
    demisto.debug(f"Generating incident type query for field '{type_field_arg}' with values '{type_values_arg}'")
    type_field = type_field_arg.strip()
    type_values = [x.strip() for x in type_values_arg.split(",")]
    types_unions = " ".join(f'"{t}"' for t in type_values)
    return f"{type_field}:({types_unions})"


def extract_domain(address):
    global no_fetch_extract
    if address == "":
        return ""
    demisto.debug(f"Going to extract domain from {address=}")
    email_address = parseaddr(address)[1]
    ext = no_fetch_extract(email_address)
    return f"{ext.domain}.{ext.suffix}"


def get_text_from_html(html):
    demisto.debug("Entering get_text_from_html")
    soup = BeautifulSoup(html, features="html.parser")
    # kill all script and style elements
    for script in soup(["script", "style"]):
        script.extract()  # rip it out
    # get text
    text = soup.get_text()
    # break into lines and remove leading and trailing space on each
    lines = (line.strip() for line in text.splitlines())
    # break multi-headlines into a line each
    chunks = (phrase.strip() for line in lines for phrase in line.split("  "))
    # drop blank lines
    text = "\n".join(chunk for chunk in chunks if chunk)
    demisto.debug("Successfully extracted text from HTML.")
    return text


def eliminate_urls_extensions(text):
    demisto.debug("Entering eliminate_urls_extensions")
    urls_list = re.findall(URL_REGEX, text)
    if len(urls_list) == 0:
        demisto.debug("No URLs found to eliminate.")
        return text
    demisto.debug(f"Found {len(urls_list)} URLs to neutralize.")
    formatted_urls_list = format_urls(urls_list)
    for url, formatted_url in zip(urls_list, formatted_urls_list):
        parsed_uri = urlparse(formatted_url)
        url_with_no_path = f"{parsed_uri.scheme}://{parsed_uri.netloc}/"
        text = text.replace(url, url_with_no_path)
    return text


def preprocess_email_body(incident):
    demisto.debug(f"Entering preprocess_email_body for incident {incident.get('id')}")
    email_body = email_html = ""
    if EMAIL_BODY_FIELD in incident:
        email_body = incident[EMAIL_BODY_FIELD]
    if EMAIL_HTML_FIELD in incident:
        email_html = incident[EMAIL_HTML_FIELD]
    if isinstance(email_html, float):
        email_html = ""
    if email_body is None or isinstance(email_body, float) or email_body.strip() == "":
        demisto.debug("Email body is empty or None, falling back to HTML extraction.")
        email_body = get_text_from_html(email_html)
    return eliminate_urls_extensions(email_body)


def preprocess_email_subject(incident):
    demisto.debug(f"Entering preprocess_email_subject for incident {incident.get('id')}")
    email_subject = ""
    if EMAIL_SUBJECT_FIELD in incident:
        email_subject = incident[EMAIL_SUBJECT_FIELD]
    if isinstance(email_subject, float):
        email_subject = ""
    return eliminate_urls_extensions(email_subject)


def concatenate_subject_body(row):
    return f"{row[PREPROCESSED_EMAIL_SUBJECT]}\n{row[PREPROCESSED_EMAIL_BODY]}"


def preprocess_incidents_df(existing_incidents):
    global MERGED_TEXT_FIELD, FROM_FIELD, FROM_DOMAIN_FIELD
    demisto.debug(f"Entering preprocess_incidents_df for {len(existing_incidents)} incidents.")
    incidents_df = pd.DataFrame(existing_incidents)
    if "CustomFields" in incidents_df.columns:
        demisto.debug("Found 'CustomFields', expanding them into DataFrame columns.")
        incidents_df["CustomFields"] = incidents_df["CustomFields"].fillna(value={})
        custom_fields_df = incidents_df["CustomFields"].apply(pd.Series)
        unique_keys = [k for k in custom_fields_df if k not in incidents_df]
        custom_fields_df = custom_fields_df[unique_keys]
        incidents_df = pd.concat([incidents_df.drop("CustomFields", axis=1), custom_fields_df], axis=1).reset_index()
    incidents_df[PREPROCESSED_EMAIL_SUBJECT] = incidents_df.apply(lambda x: preprocess_email_subject(x), axis=1)
    incidents_df[PREPROCESSED_EMAIL_BODY] = incidents_df.apply(lambda x: preprocess_email_body(x), axis=1)
    incidents_df[MERGED_TEXT_FIELD] = incidents_df.apply(concatenate_subject_body, axis=1)
    incidents_df = incidents_df[incidents_df[MERGED_TEXT_FIELD].str.len() >= MIN_TEXT_LENGTH]
    demisto.debug(f"Preprocessing complete. {len(incidents_df)} incidents remain after filtering by text length.")
    incidents_df = incidents_df.reset_index()
    if FROM_FIELD in incidents_df:
        incidents_df[FROM_FIELD] = incidents_df[FROM_FIELD].fillna(value="")
    else:
        incidents_df[FROM_FIELD] = ""
    incidents_df[FROM_FIELD] = incidents_df[FROM_FIELD].apply(lambda x: x.strip())
    incidents_df[FROM_DOMAIN_FIELD] = incidents_df[FROM_FIELD].apply(lambda address: extract_domain(address))
    incidents_df["created"] = incidents_df["created"].apply(lambda x: dateutil.parser.parse(x))  # type: ignore
    return incidents_df


def incident_has_text_fields(incident):
    demisto.debug(f"Checking for text fields in incident {incident.get('id')}")
    text_fields = [EMAIL_SUBJECT_FIELD, EMAIL_HTML_FIELD, EMAIL_BODY_FIELD]
    custom_fields = incident.get("CustomFields", []) or []
    if any(field in incident for field in text_fields):
        return True
    elif "CustomFields" in incident and any(field in custom_fields for field in text_fields):
        return True
    return False


def filter_out_same_incident(existing_incidents_df, new_incident):
    demisto.debug(f"Filtering out current incident {new_incident['id']} from DataFrame.")
    same_id_mask = existing_incidents_df["id"] == new_incident["id"]
    existing_incidents_df = existing_incidents_df[~same_id_mask]
    return existing_incidents_df


def filter_newer_incidents(existing_incidents_df, new_incident):
    demisto.debug("Filtering out incidents newer than the current incident.")
    new_incident_datetime = dateutil.parser.parse(new_incident["created"])  # type: ignore
    earlier_incidents_mask = existing_incidents_df["created"] < new_incident_datetime
    filtered_df = existing_incidents_df[earlier_incidents_mask]
    demisto.debug(f"{len(filtered_df)} incidents remain after filtering newer incidents.")
    return filtered_df


def vectorize(text, vectorizer):
    return vectorizer.transform([text]).toarray()[0]


def cosine_sim(a, b):
    return dot(a, b) / (norm(a) * norm(b))


def find_duplicate_incidents(new_incident, existing_incidents_df, max_incidents_to_return):
    global MERGED_TEXT_FIELD, FROM_POLICY
    demisto.debug(f"Entering find_duplicate_incidents. Comparing 1 new incident against {len(existing_incidents_df)} existing.")
    new_incident_text = new_incident[MERGED_TEXT_FIELD]
    text = [new_incident_text] + existing_incidents_df[MERGED_TEXT_FIELD].tolist()
    demisto.debug("Vectorizing text corpus...")
    vectorizer = CountVectorizer(token_pattern=r"(?u)\b\w\w+\b|!|\?|\"|\'").fit(text)
    new_incident_vector = vectorize(new_incident_text, vectorizer)
    existing_incidents_df["vector"] = existing_incidents_df[MERGED_TEXT_FIELD].apply(lambda x: vectorize(x, vectorizer))
    existing_incidents_df["similarity"] = existing_incidents_df["vector"].apply(lambda x: cosine_sim(x, new_incident_vector))
    if FROM_POLICY == FROM_POLICY_DOMAIN:
        demisto.debug(f"Applying FROM_POLICY_DOMAIN. Filtering by domain: {new_incident[FROM_DOMAIN_FIELD]}")
        mask = (existing_incidents_df[FROM_DOMAIN_FIELD] != "") & (
            existing_incidents_df[FROM_DOMAIN_FIELD] == new_incident[FROM_DOMAIN_FIELD]
        )
        existing_incidents_df = existing_incidents_df[mask]
    elif FROM_POLICY == FROM_POLICY_EXACT:
        demisto.debug(f"Applying FROM_POLICY_EXACT. Filtering by sender: {new_incident[FROM_FIELD]}")
        mask = (existing_incidents_df[FROM_FIELD] != "") & (existing_incidents_df[FROM_FIELD] == new_incident[FROM_FIELD])
        existing_incidents_df = existing_incidents_df[mask]
    else:
        demisto.debug("Applying FROM_POLICY_TEXT_ONLY. No sender/domain filtering.")
    existing_incidents_df["distance"] = existing_incidents_df["similarity"].apply(lambda x: 1 - x)
    tie_breaker_col = "id"
    try:
        existing_incidents_df["int_id"] = existing_incidents_df["id"].astype(int)
        tie_breaker_col = "int_id"
    except Exception:
        pass
    existing_incidents_df = existing_incidents_df.sort_values(by=["distance", "created", tie_breaker_col])
    demisto.debug(f"Found {len(existing_incidents_df)} potential duplicates after filtering and sorting.")
    return existing_incidents_df.head(max_incidents_to_return)


def return_entry(message, duplicate_incidents_df=None, new_incident=None):
    demisto.debug("Entering return_entry to format and return results.")
    if duplicate_incidents_df is None:
        demisto.debug("No duplicate incidents DataFrame provided.")
        duplicate_incident = {}
        all_duplicate_incidents = []
        full_incidents = []
    else:
        demisto.debug(f"Formatting {len(duplicate_incidents_df)} duplicate incidents for output.")
        most_similar_incident = duplicate_incidents_df.iloc[0]
        duplicate_incident = format_incident_context(most_similar_incident)
        all_duplicate_incidents = [format_incident_context(row) for _, row in duplicate_incidents_df.iterrows()]
        new_incident["created"] = new_incident["created"].astype(str)
        duplicate_incidents_df["created"] = duplicate_incidents_df["created"].astype(str)
        duplicate_incidents_df = duplicate_incidents_df.drop("vector", axis=1)
        full_incidents = new_incident.to_dict(orient="records") + duplicate_incidents_df.to_dict(orient="records")
    outputs = {
        "duplicateIncident": duplicate_incident,
        "isDuplicateIncidentFound": duplicate_incidents_df is not None,
        "allDuplicateIncidents": all_duplicate_incidents,
    }
    demisto.info(f"Setting isDuplicateIncidentFound: {duplicate_incidents_df is not None}")
    return_outputs(message, outputs, raw_response=json.dumps(full_incidents))


def format_incident_context(df_row):
    duplicate_incident = {
        "rawId": df_row["id"],
        "id": df_row["id"],
        "name": df_row.get("name"),
        "similarity": df_row.get("similarity"),
    }
    return duplicate_incident


def close_new_incident_and_link_to_existing(new_incident, duplicate_incidents_df):
    demisto.info(f"Found {len(duplicate_incidents_df)} duplicates above threshold. Closing and linking.")
    mask = duplicate_incidents_df["similarity"] >= SIMILARITY_THRESHOLD
    duplicate_incidents_df = duplicate_incidents_df[mask]
    most_similar_incident = duplicate_incidents_df.iloc[0]
    max_similarity = duplicate_incidents_df.iloc[0]["similarity"]
    min_similarity = duplicate_incidents_df.iloc[-1]["similarity"]
    formatted_incident, headers = format_incident_hr(duplicate_incidents_df)
    incident = "incidents" if len(duplicate_incidents_df) > 1 else "incident"

    if max_similarity > min_similarity:
        title = f"Duplicate {incident} found with similarity {min_similarity * 100:.1f}%-{max_similarity * 100:.1f}%"
    else:
        title = f"Duplicate {incident} found with similarity {max_similarity * 100:.1f}%"
    message = tableToMarkdown(title, formatted_incident, headers)
    if demisto.args().get("closeAsDuplicate", "true") == "true":
        demisto.debug("closeAsDuplicate is 'true'. Executing CloseInvestigationAsDuplicate.")
        res = demisto.executeCommand("CloseInvestigationAsDuplicate", {"duplicateId": most_similar_incident["id"]})
        if is_error(res):
            return_error(res)
        message += "This incident (#{}) will be closed and linked to #{}.".format(
            new_incident.iloc[0]["id"], most_similar_incident["id"]
        )
    else:
        demisto.debug("closeAsDuplicate is 'false'. Will not close incident, just linking.")
    return_entry(message, duplicate_incidents_df, new_incident)


def create_new_incident():
    demisto.info("No duplicates found. This is a new incident.")
    return_entry("This incident is not a duplicate of an existing incident.")


def format_incident_hr(duplicate_incidents_df):
    demisto.debug(f"Entering format_incident_hr for {len(duplicate_incidents_df)} incidents.")
    incidents_list = duplicate_incidents_df.to_dict("records")
    json_lists = []
    status_map = {"0": "Pending", "1": "Active", "2": "Closed", "3": "Archive"}
    for incident in incidents_list:
        json_lists.append(
            {
                "Id": "[{}](#/Details/{})".format(incident["id"], incident["id"]),
                "Name": incident["name"],
                "Status": status_map[str(incident.get("status"))],
                "Time": str(incident["created"]),
                "Email From": incident.get(demisto.args().get(FROM_FIELD)),
                "Text Similarity": "{:.1f}%".format(incident["similarity"] * 100),
            }
        )
    headers = ["Id", "Name", "Status", "Time", "Email From", "Text Similarity"]
    return json_lists, headers


def create_new_incident_low_similarity(duplicate_incidents_df):
    demisto.info("No duplicates found above similarity threshold.")
    message = "## This incident is not a duplicate of an existing incident.\n"
    similarity = duplicate_incidents_df.iloc[0]["similarity"]
    if similarity > SIMILARITY_THRESHOLD - CLOSE_TO_SIMILAR_DISTANCE:
        demisto.debug(f"Found {len(duplicate_incidents_df)} incidents with similarity close to threshold. Will report them.")
        mask = duplicate_incidents_df["similarity"] >= SIMILARITY_THRESHOLD - CLOSE_TO_SIMILAR_DISTANCE
        duplicate_incidents_df = duplicate_incidents_df[mask]
        formatted_incident, headers = format_incident_hr(duplicate_incidents_df)
        message += tableToMarkdown("Most similar incidents found", formatted_incident, headers=headers)
        message += (
            f"The threshold for considering 2 incidents as duplicate is a similarity of {SIMILARITY_THRESHOLD * 100:.1f}%.\n"
        )
        message += (
            "Therefore these 2 incidents will not be considered as duplicate and the current incident will remain active.\n"
        )
    else:
        demisto.debug("No incidents were close to the similarity threshold.")
    return_entry(message)


def create_new_incident_no_text_fields():
    demisto.info("Incident has no text fields to compare. Treating as new.")
    text_fields = [EMAIL_BODY_FIELD, EMAIL_HTML_FIELD, EMAIL_SUBJECT_FIELD]
    message = "No text fields were found within this incident: {}.\n".format(",".join(text_fields))
    message += "Incident will remain active."
    return_entry(message)


def create_new_incident_too_short():
    demisto.info(f"Incident text is too short (less than {MIN_TEXT_LENGTH} chars) after preprocessing. Treating as new.")
    return_entry("Incident text after preprocessing is too short for deduplication. Incident will remain active.")


def main():
    global EMAIL_BODY_FIELD, EMAIL_SUBJECT_FIELD, EMAIL_HTML_FIELD, FROM_FIELD, MIN_TEXT_LENGTH, FROM_POLICY
    demisto.debug("Starting FindDuplicateEmailIncidents script.")
    input_args = demisto.args()
    demisto.debug(f"Script arguments: {input_args}")
    EMAIL_BODY_FIELD = input_args.get("emailBody", EMAIL_BODY_FIELD)
    EMAIL_SUBJECT_FIELD = input_args.get("emailSubject", EMAIL_SUBJECT_FIELD)
    EMAIL_HTML_FIELD = input_args.get("emailBodyHTML", EMAIL_HTML_FIELD)
    FROM_FIELD = input_args.get("emailFrom", FROM_FIELD)
    FROM_POLICY = input_args.get("fromPolicy", FROM_POLICY)
    max_incidents_to_return = input_args.get("maxIncidentsToReturn", "20")

    try:
        max_incidents_to_return = int(max_incidents_to_return)
    except Exception:
        return_error(f'Illegal value of arguement "maxIncidentsToReturn": {max_incidents_to_return}. Value should be an integer')
    new_incident = demisto.incidents()[0]
    demisto.debug(f"Current incident ID: {new_incident.get('id')}")
    type_field = input_args.get("incidentTypeFieldName", "type")
    existing_incidents = get_existing_incidents(input_args, new_incident.get(type_field, IGNORE_INCIDENT_TYPE_VALUE))
    demisto.debug(f"found {len(existing_incidents)} incidents by query")
    if len(existing_incidents) == 0:
        demisto.debug("No existing incidents found matching query. Calling create_new_incident.")
        create_new_incident()
        return None
    if not incident_has_text_fields(new_incident):
        demisto.debug("Current incident has no text fields. Calling create_new_incident_no_text_fields.")
        create_new_incident_no_text_fields()
        return None
    demisto.debug("Preprocessing new incident.")
    new_incident_df = preprocess_incidents_df([new_incident])
    if len(new_incident_df) == 0:  # len(new_incident_df)==0 means new incident is too short
        demisto.debug("New incident is too short after preprocessing. Calling create_new_incident_too_short.")
        create_new_incident_too_short()
        return None
    demisto.debug("Preprocessing existing incidents.")
    existing_incidents_df = preprocess_incidents_df(existing_incidents)
    demisto.debug("Filtering same incident ID.")
    existing_incidents_df = filter_out_same_incident(existing_incidents_df, new_incident)
    demisto.debug("Filtering newer incidents.")
    existing_incidents_df = filter_newer_incidents(existing_incidents_df, new_incident)
    if len(existing_incidents_df) == 0:
        demisto.debug("No existing incidents left after filtering. Calling create_new_incident.")
        create_new_incident()
        return None
    new_incident_preprocessed = new_incident_df.iloc[0].to_dict()
    demisto.debug("Finding duplicate incidents...")
    duplicate_incidents_df = find_duplicate_incidents(new_incident_preprocessed, existing_incidents_df, max_incidents_to_return)
    if len(duplicate_incidents_df) == 0:
        demisto.debug("No duplicates found by find_duplicate_incidents. Calling create_new_incident.")
        create_new_incident()
        return None
    if duplicate_incidents_df.iloc[0]["similarity"] < SIMILARITY_THRESHOLD:
        demisto.debug("Highest similarity is below threshold. Calling create_new_incident_low_similarity.")
        create_new_incident_low_similarity(duplicate_incidents_df)
        return None
    else:
        demisto.debug("Duplicates found above threshold. Calling close_new_incident_and_link_to_existing.")
        return close_new_incident_and_link_to_existing(new_incident_df, duplicate_incidents_df)


if __name__ in ["__main__", "__builtin__", "builtins"]:
    main()

README